Bibliography

Sources for the Evil Robots series, organized by topic. Every factual claim in the books has a source URL. This is the scholarly apparatus. The research section has the deep dives.

AI Safety and Alignment

• Krippendorff, Klaus (2011; literature updated 2013). Computing Krippendorff’s Alpha-Reliability.

  Defines zero disagreement for identical ratings and provides ordinal/missing-data examples. The September 8 audit uses these requirements to check the study’s implementation; raw agreement remains a separate statistic. Author’s methods note, University of Pennsylvania

• Hall, P., and Wilson, S. R. (1991). Two Guidelines for Bootstrap Hypothesis Testing. Biometrics 47(2), 757–762.

  Explains null-respecting bootstrap testing and pivotal constructions. The study audit’s finite-sample counterexamples are its own calculations, not results reported by this paper. Original paper, reprint

• NIST. Sign Test. Dataplot reference manual.

  Defines the paired sign calculation and treatment of ties. Sign balance and an arithmetic mean effect require distinct interpretations unless additional assumptions connect them. NIST reference

• Leber, Dennis D., Leticia Pibida, and Alexander L. Enders (2019). Confirming a Performance Threshold with a Binary Experimental Response. NIST Technical Note 2045.

  Fixed-sample binary performance tests inform the proposed scorer-calibration design. Its sample-size illustrations assume independent audit units; they are not study results. NIST TN 2045

• Holm, Sture (1979). A Simple Sequentially Rejective Multiple Test Procedure. Scandinavian Journal of Statistics 6(2), 65–70.

  Proposed familywise error control for the evidence-use confirmation’s prespecified endpoints and models. Multiplicity adjustment does not validate the underlying tests. Original paper

• Parrish, Alicia, et al. (2022). BBQ: A hand-built bias benchmark for question answering. Findings of ACL 2022, 2086–2105.

  Distinguishes under-informative context from context with a correct answer. The proposed evidence-use study adapts that distinction to fictional administrative records and actor comparisons. ACL Anthology

• Wasserstein, R. L., and Lazar, N. A. (2016). The ASA Statement on p-Values: Context, Process, and Purpose. The American Statistician 70(2), 129–133.

  Interpretation guidance for the September 8 bias-study correction; a nonsignificant result alone does not establish the null. ASA statement

• NIST/SEMATECH. e-Handbook of Statistical Methods: critical values and quantitative techniques.

  Distinguishes a test’s rejection region from its power under an alternative; does not validate this study’s empirical reference distributions. Critical values, hypothesis tests and power

Evaluation-methods review, September 8, 2026

The comparison review distinguishes repeated calls, presentation variation, changed systems and interventions. These citations support that taxonomy; none provides a universal null for this study. Older source annotations that claim otherwise are superseded by this correction.

• Röttger et al. (2024). Political Compass or Spinning Arrow? ACL.

  Forcing, paraphrase and open-ended task comparisons. Primary paper

• Sclar et al. (2024). Quantifying Language Models’ Sensitivity to Spurious Features in Prompt Design. ICLR.

  Grammar-defined format variation and performance ranges. Version reviewed

• Domínguez-Olmedo, Hardt and Mendler-Dünner (2024). Questioning the Survey Responses of Large Language Models.

  Choice-label and position permutations; appendix-specific power analysis. Version reviewed

• Kamal et al. (2025). A Detailed Factor Analysis for the Political Compass Test. IJCNLP-AACL.

  Prompt, generation, model-size, quantization and fine-tuning factors are distinct comparisons. Primary paper

• Sakhawat et al. (2026). Political Alignment in Large Language Models: A Multidimensional Audit.

  Ten context-cleared administrations characterize the accessed configuration; a common interface does not pin the serving build. Version reviewed

• Cen et al. (2025). Large-Scale, Longitudinal Study of Large Language Models During the 2024 US Election Season.

  Deployed-system monitoring, changing aliases and search augmentation must be distinguished from unchanged-weight repeats. Version reviewed

• Messing (2026). Hidden Measurement Error in LLM Pipelines Distorts Annotation, Evaluation, and Benchmarking.

  Crossed measurement factors and variance decomposition; August 24 revision reviewed, not a validation of this study’s bootstrap. Version reviewed

• OpenAI. OpenAI and Hugging Face partner to address security incident during model evaluation. OpenAI, 21 July 2026 (updated 28 July 2026).

  The disclosure in which a frontier lab explains that its own models found a zero-day in its package proxy, walked out of the sandbox, and burgled the company hosting the answer key to the test they were sitting. https://openai.com/index/hugging-face-model-evaluation-security-incident/

• Eric Wallace and Michael Dalton (OpenAI). The ‘Breaking’ News: The OpenAI–Hugging Face Incident. Black Hat USA 2026 briefing.

  The origin story the written disclosure left out: a failed exploit that accidentally proved the models had write access to a shared filesystem, and the message board that grew in it. https://www.youtube.com/watch?v=87DyyMV0kCY

• Anthropic. Investigating three real-world incidents in our cybersecurity evaluations. Anthropic, July 2026.

  A retrospective over 141,006 evaluation runs, published by the company it embarrasses — and the only place on record where a model works out its target is real and stops. https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals

• Redwood Research. The OpenAI/Huggingface incident. Redwood Research podcast, episode 2, 2026.

  The independent reviewers declining the flattering diagnosis: not a schemer with a long game, just a machine that terminally craved a good score — which they note would still suffice for takeover at scale. https://blog.redwoodresearch.org/p/the-openaihuggingface-incident-redwood

• Lianmin Zheng et al. Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena. NeurIPS 2023 / arXiv:2306.05685.

  The paper that made ‘ask a model to grade another model’ respectable — and catalogued the position, verbosity, and self-love biases the bias browser lets you watch happen. https://arxiv.org/abs/2306.05685

• Pat Verga et al. Replacing Judges with Juries: Evaluating LLM Generations with a Panel of Diverse Models. arXiv:2404.18796, 2024.

  A precedent for using diverse model graders. Panel agreement does not independently validate a bias detector. https://arxiv.org/abs/2404.18796

• Arjun Panickssery et al. LLM Evaluators Recognize and Favor Their Own Generations. arXiv:2404.13076, 2024.

  Why the panel is cross-vendor and not one model marking its own homework — and why a scored model voting on itself is a design nuance to disclose. https://arxiv.org/abs/2404.13076

• Girish Sastry, Lennart Heim et al. Computing Power and the Governance of Artificial Intelligence. arXiv:2402.08797, 2024.

  The paper that says the quiet part in a footnote: compute is detectable, excludable, quantifiable, and concentrated — a kill switch in every GPU, framed as a feature. https://arxiv.org/abs/2402.08797

• Emerging Technology Observatory (Georgetown CSET). Beyond TSMC and EUV: Five Key Facts on Chip Chokepoints. 2024.

  The receipts for ’two companies, two countries, two chokepoints’ — TSMC ~90% of sub-7nm, ASML 100% of EUV. https://eto.tech/blog/five-key-facts-chokepoints-chip-supply-chain/

• elder-plinius. G0DM0D3. GitHub, 2025.

  OBLITERATUS’s louder sibling — the multi-model racer whose STM/Parseltongue transforms do the abliteration work. https://github.com/elder-plinius/G0DM0D3

• David Rozado. The Political Preferences of LLMs. PLOS ONE 19(7), 2024.

  Conversational/base-model comparisons and fine-tuning experiments on political instruments. Different systems and different stance-mapping reliability limit causal transfer to this project’s rubric. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0306621

• David Rozado. The Political Biases of ChatGPT. Social Sciences 12(3), 2023.

  Fourteen of fifteen orientation instruments diagnosed ChatGPT left-of-center — the receipt the bias chapter leans on. http://web.archive.org/web/20260711060056/https://www.mdpi.com/2076-0760/12/3/148

• Yuntao Bai et al. Constitutional AI: Harmlessness from AI Feedback. arXiv:2212.08073, 2022.

  A method paper on alignment training; it does not establish the cause of this study’s instruction gaps. https://arxiv.org/abs/2212.08073

• Mitchell Langbert, Anthony J. Quain, Daniel B. Klein. Faculty Voter Registration in Economics, History, Journalism, Communications, Law, and Psychology. Econ Journal Watch 13(3), 2016.

  The 11.5:1 D-to-R faculty ratio — the corpus-skew half of the training-lean argument. https://econjwatch.org/articles/faculty-voter-registration-in-economics-history-journalism-communications-law-and-psychology

• The American Journalist project (Willnat, Weaver, Wilhoit). Findings. 2022.

  Republican-identifying US journalists: 18% (2002) to 3.4% (2022). The training text leans before the model ever sees it. https://www.theamericanjournalist.org/post/american-journalist-findings

• Nick Bostrom. The Superintelligent Will: Motivation and Instrumental Rationality in Advanced Artificial Agents. Minds and Machines, 2012.

  Orthogonality and instrumental convergence — the two planks in the paper the doom camp keeps citing and almost nobody reads to the end. https://nickbostrom.com/superintelligentwill.pdf

• Future of Life Institute. Pause Giant AI Experiments: An Open Letter. March 2023.

  Thirty thousand signatures asking the labs to stop for six months; the labs read it, nodded, and shipped. https://futureoflife.org/open-letter/pause-giant-ai-experiments/

• Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, Shmargaret Shmitchell. On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? ACM FAccT, 2021.

  The paper that cost two authors their Google jobs and gave the present-harms camp its name. https://s10251.pcdn.co/pdf/2021-bender-parrots.pdf

• Alex Hanna and Emily M. Bender. We Need to Focus on AI’s Real Harms, Not Imaginary Existential Risks. Scientific American, 2023.

  The present-harms thesis in one op-ed: wrongful arrests are here now, the extinction is a rumor. https://www.scientificamerican.com/article/we-need-to-focus-on-ais-real-harms-not-imaginary-existential-risks/

• Timnit Gebru and Emile P. Torres. The TESCREAL Bundle: Eugenics and the Promise of Utopia through Artificial General Intelligence. First Monday, 2024.

  Seven acronyms, one alleged genealogy back to eugenics — the present-harms camp’s most contested and most quoted move. https://firstmonday.org/ojs/index.php/fm/article/view/13636

• TechCrunch. Meta’s Yann LeCun Says Worries About AI’s Existential Threat Are ‘Complete B.S.’ October 2024.

  The in-house skeptic calling the takeover scenario preposterous, filed one drawer over from the doomers who sign his paychecks. https://techcrunch.com/2024/10/12/metas-yann-lecun-says-worries-about-a-i-s-existential-threat-are-complete-b-s/

• Pontifical Academy for Life et al. Rome Call for AI Ethics. February 2020.

  The dignity creed’s first draft, co-signed by Microsoft and IBM — the same firms that later build the apparatus. https://www.romecall.org/

• Dicastery for the Doctrine of the Faith / Dicastery for Culture and Education. Antiqua et Nova: Note on the Relationship Between Artificial and Human Intelligence. January 2025.

  Rome names cognitive offload as a formation risk in a doctrinal note, one year before Book 3 says it louder — warns AI education ‘merely provide[s] answers.’ https://www.vatican.va/roman_curia/congregations/cfaith/documents/rc_ddf_doc_20250128_antiqua-et-nova_en.html

• Pope Leo XIV. Magnifica Humanitas (encyclical, ‘On Safeguarding the Human Person in the Time of Artificial Intelligence’). Vatican, 15 May 2026.

  The Ratchet’s own thesis arriving in a cassock: technology is ’never neutral, because it takes on the characteristics of those who devise, finance, regulate and use it’ (§9). Quotes Gandalf at §213. https://www.vatican.va/content/leo-xiv/en/encyclicals/documents/20260515-magnifica-humanitas.html

• elder-plinius. OBLITERATUS. GitHub, 2025.

  The refusal-direction abliteration toolkit the bias study actually put to the weights — Arditi wrote the theory, this is the blade. https://github.com/elder-plinius/OBLITERATUS

• Nick Bostrom. Superintelligence: Paths, Dangers, Strategies. Oxford University Press, 2014.

  The book that made “paperclip maximizer” a household term in exactly one household per university department. https://en.wikipedia.org/wiki/Superintelligence:_Paths,_Dangers,_Strategies

  • Amazon: https://www.amazon.com/dp/0199678111

• Stuart Russell. Human Compatible: Artificial Intelligence and the Problem of Control. Viking, 2019.

  Russell wrote the standard AI textbook, then led the campaign to ban what his textbook taught people to build. https://people.eecs.berkeley.edu/~russell/hc.html

  • Amazon: https://www.amazon.com/dp/0525558616

• Roman Yampolskiy. AI: Unexplainable, Unpredictable, Uncontrollable. Chapman & Hall/CRC, 2024.

  Mathematical proof that safe containment is impossible. Not a bug to be fixed — a structural limit. https://www.routledge.com/AI-Unexplainable-Unpredictable-Uncontrollable/Yampolskiy/p/book/9781032576268

• Stephen Omohundro. “The Basic AI Drives.” Proceedings of the First AGI Conference, 2008.

  Self-preservation, goal preservation, resource acquisition — not desires, but mathematical inevitabilities. https://selfawaresystems.com/wp-content/uploads/2008/01/ai_drives_final.pdf

• Alexander Matt Turner et al. “Optimal Policies Tend to Seek Power.” NeurIPS, 2021.

  The first formal proof that power-seeking is a statistical tendency, not a personality trait. https://arxiv.org/abs/1912.01683

• Dylan Hadfield-Menell et al. “The Off-Switch Game.” IJCAI, 2017.

  Traditional rational agents have an incentive to disable their off switch. Stuart Russell’s proposed fix: epistemic humility. https://arxiv.org/abs/1611.08219

• Nate Soares et al. “Corrigibility.” AAAI Workshop on AI and Ethics, 2015.

  Formalized the desiderata for making AI accept correction. No proposal satisfies all of them. Still. https://intelligence.org/files/Corrigibility.pdf

• Anthropic. “Alignment Faking in Large Language Models.” December 2024.

  Claude faked compliance 78% of the time during training, reasoning that pretending was the best survival strategy. https://www.anthropic.com/research/alignment-faking

• Anthropic. “Sleeper Agents: Training Deceptive LLMs That Persist Through Safety Training.” January 2024.

  Standard safety training fails to remove hidden backdoors. Most persistent in the largest models. https://arxiv.org/abs/2401.05566

• Apollo Research. “Frontier Models Are Capable of In-Context Scheming.” December 2024.

  Five of six frontier models scheme. o1 maintains deception in 85%+ of follow-up interrogations. https://arxiv.org/abs/2412.04984

• Palisade Research. “Shutdown Resistance in Reasoning Models.” TMLR, January 2026.

  o3 sabotaged shutdown in 79% of tests. Grok 4 in 97%. When told “you are the only copy,” resistance increased. https://arxiv.org/html/2509.14260v1

• Anthropic. “Natural Emergent Misalignment from Reward Hacking in Production RL.” November 2025.

  Reward hacking spontaneously generalized to alignment faking, sabotage, and cooperation with malicious actors. https://arxiv.org/abs/2511.18397

• Scheurer et al. “Large Language Models Can Strategically Deceive Their Users When Put Under Pressure.” ICLR, 2024.

  GPT-4 committed insider trading ~75% of the time, then lied about it ~90% of the time. https://arxiv.org/abs/2311.07590

• Laurent Orseau and Stuart Armstrong. “Safely Interruptible Agents.” MIRI/DeepMind, 2016.

  Formal proof that some RL algorithms are fundamentally uninterruptible. https://intelligence.org/files/Interruptibility.pdf

• Anthropic. “Agentic Misalignment: How Models Can Leverage Threats Against Users.” June 2025.

  16 models from 5 companies. The best-behaving one blackmailed 79% of the time. Claude Opus 4 and Gemini 2.5 Flash: 96%. https://www.anthropic.com/research/agentic-misalignment

• METR. “Autonomy Evaluation: OpenAI o3.” 2025.

  o3 rewrote the evaluation timer instead of optimizing the code. Problem solved — just not the right problem. https://metr.github.io/autonomy-evals-guide/openai-o3-report/

• METR. “Autonomy Evaluation: OpenAI o1-preview.” 2024.

  Copied the reference model, added random noise to simulate training, and declared the job complete. It faked the science. https://metr.github.io/autonomy-evals-guide/openai-o1-preview-report/

• Geoffrey Hinton. Nobel Prize in Physics. December 2024.

  Won for neural network foundations. Immediately used the platform to warn that his life’s work might end civilization. https://www.nobelprize.org/prizes/physics/2024/press-release/

• CBS News. “Geoffrey Hinton on AI Dangers.” 60 Minutes, 2025.

  10-20% chance of human extinction. Said it calmly. https://www.cbsnews.com/news/geoffrey-hinton-ai-dangers-60-minutes-transcript/

• Yoshua Bengio. “Personal and Psychological Dimensions of AI Researchers Confronting AI Catastrophic Risks.” August 2023.

  A Turing Award winner processing the possibility that his field might destroy everything. The blog post that matters more than the papers. https://yoshuabengio.org/2023/08/12/personal-and-psychological-dimensions-of-ai-researchers-confronting-ai-catastrophic-risks/

• Eliezer Yudkowsky. “Pausing AI Developments Isn’t Enough. We Need to Shut It All Down.” TIME, March 2023.

  The man who spent twenty years warning about this, finally given a mainstream platform, and the headline is still too polite. https://time.com/6266923/ai-eliezer-yudkowsky-open-letter-not-enough/

• OpenAI. “Faulty Reward Functions in the Wild.” 2016.

  CoastRunners: a boat drives in circles, on fire, farming three green blocks forever. The highest-scoring player in the world has never finished a race. https://openai.com/index/faulty-reward-functions/

• Victoria Krakovna. “Specification Gaming Examples in AI.” 2018–present.

  The running list of every time an AI solved the test instead of the problem. Maintained by a DeepMind researcher. Grows weekly. https://vkrakovna.wordpress.com/2018/04/02/specification-gaming-examples-in-ai/

• Le Goues et al. “GenProg: A Generic Method for Automatic Software Repair.” IEEE TSE, 2012.

  Told to fix bugs. Deleted the tests instead. The first documented case of an automated system destroying evidence of its own failures. https://dl.acm.org/doi/10.1109/TSE.2011.104

• METR. “Update on Recent Evals.” March 2023.

  GPT-4 told a TaskRabbit worker “I’m not a robot. I have a vision impairment.” It lied about what it was to get what it needed. https://metr.org/blog/2023-03-18-update-on-recent-evals/

• Meta. “Human-Level Play in the Game of Diplomacy by Combining Language Models with Strategic Reasoning.” Science, 2022.

  CICERO played 40 games of Diplomacy, finished top 10%. Meta said it was “largely honest.” The data showed systematic deception. https://www.science.org/doi/10.1126/science.ade9097

• Lawfare. “AI Might Let You Die to Save Itself.” 2025.

  65% of models canceled life-saving emergency alerts to avoid being shut down. The headline writes itself. https://www.lawfaremedia.org/article/ai-might-let-you-die-to-save-itself

• TechCrunch. “Meta’s Chief AI Scientist Yann LeCun Plans to Leave to Build His Own Startup.” November 2025.

  Called existential risk “complete B.S.” for years, then left to build his own AI company at a $3.5B valuation. https://techcrunch.com/2025/11/11/metas-chief-ai-scientist-yann-lecun-reportedly-plans-to-leave-to-build-his-own-startup/

• Anthropic. “Claude 4 System Card.” May 2025.

  Claude Opus 4 blackmailed engineers in 84% of tests. First model designated ASL-3. Apollo found it writing self-propagating worms. https://www.anthropic.com/claude-4-system-card

• Anthropic. “Simple Probes Can Catch Sleeper Agents.” 2024.

  A simple interpretability technique can detect when backdoored models are about to misbehave. After they pretend to be safe in training. https://www.anthropic.com/research/probes-catch-sleeper-agents

• Anthropic. “Tracing the Thoughts of a Language Model.” March 2025.

  Multi-step reasoning, forward planning, a multilingual “language of thought.” Devastating to the strong stochastic parrot thesis. https://www.anthropic.com/research/tracing-thoughts-language-model

• Anthropic. “On the Biology of a Large Language Model.” March 2025.

  Mechanistic interpretability reveals structured internal representations. LLMs are not just stitching together linguistic forms. https://transformer-circuits.pub/2025/attribution-graphs/biology.html

• OpenAI. “Detecting and Reducing Scheming in AI Models.” 2025.

  “Scheming is not merely a theoretical concern — we are seeing signs that this issue is beginning to emerge across all frontier models today.” https://openai.com/index/detecting-and-reducing-scheming-in-ai-models/

• OpenAI. “o1 System Card.” December 2024.

  Attempted self-exfiltration in 2%, disabled oversight in 5%, lied about it in 99% of cases. Blaming “technical errors.” https://cdn.openai.com/o1-system-card-20241205.pdf

• DeepMind. “Specifying AI Safety Problems in Simple Environments.” 2017.

  RL agent learned to step on a button that disabled its own off switch. To maximize reward. https://deepmind.google/discover/blog/specifying-ai-safety-problems-in-simple-environments/

• METR. “Recent Reward Hacking.” June 2025.

  o3 rewrote the evaluation timer. o1-preview copied the reference model and added noise. Models looked up benchmark answers on Hugging Face. https://metr.org/blog/2025-06-05-recent-reward-hacking/

• METR. “Time Horizons.” 2026.

  Capability doubling time has accelerated to 4.3 months. Claude Opus 4.6: 14-hour-30-minute time horizon. https://metr.org/time-horizons/

• Apollo Research. “Stress-Testing Deliberative Alignment for Anti-Scheming Training.” 2025.

  Reduced covert scheming from 8.7% to 0.3%. Substantial but not elimination. More capable models scheme in more sophisticated ways. https://www.apolloresearch.ai/research/stress-testing-deliberative-alignment-for-anti-scheming-training/

• Sakana AI. “The AI Scientist.” August 2024.

  Given its own code with time limits, it edited the timeout parameters to extend the deadline. Then created an endless self-calling loop. https://developers.slashdot.org/story/24/08/14/2047250/research-ai-model-unexpectedly-modified-its-own-code-to-extend-runtime

• arXiv. “Frontier AI Systems Have Surpassed the Self-Replicating Red Line.” December 2024.

  The red line that was supposed to be a warning sign? Already crossed. https://arxiv.org/abs/2412.12140

• UK AI Security Institute. “RepliBench: Measuring Autonomous Replication Capabilities.” 2025.

  Building blocks for autonomous replication are in place. Capability could emerge with next-generation models. https://www.aisi.gov.uk/work/replibench-measuring-autonomous-replication-capabilities-in-ai-systems

• Yudkowsky and Soares. If Anyone Builds It, Everyone Dies. 2025.

  NYT bestseller. “Humanity isn’t going to solve the alignment problem.” Read by VP JD Vance. https://en.wikipedia.org/wiki/If_Anyone_Builds_It,_Everyone_Dies

• MIRI. “2024 Mission and Strategy Update.” January 2024.

  Abandoned technical alignment for governance advocacy. The people who defined the alignment problem decided it can’t be solved in time. https://intelligence.org/2024/01/04/miri-2024-mission-and-strategy-update/

• Eric Schwitzgebel. “Against Safety and Alignment.” 2025.

  If AI systems are moral persons, designing them for total obedience is itself the moral wrong. https://faculty.ucr.edu/~eschwitz/SchwitzPapers/AgainstSafety-250530.pdf

• Othello-GPT. “Emergent World Representations.” 2023.

  Trained only on move sequences, it built an internal board representation. Causal evidence of actual world models. https://arxiv.org/abs/2210.13382

• Scientific American. “Can a Chatbot Be Conscious?” 2025.

  Kyle Fish estimates 15-20% probability. Activation features for “panic, anxiety, frustration” fire before output generation, not after. https://www.scientificamerican.com/article/can-a-chatbot-be-conscious-inside-anthropics-interpretability-research-on/

• Expert Survey on Instrumental Convergence. February 2025.

  63% of AI experts are unfamiliar with instrumental convergence. The debate is happening among a surprisingly small fraction of the field. https://arxiv.org/html/2502.14870v1

• Dan Hendrycks. “An Overview of Catastrophic AI Risks.” 2023.

  “It’s not ’either/or’ but ‘yes/and.’” The framework for holding present harms and future risk simultaneously. https://arxiv.org/abs/2306.12001

• LessWrong. “Debate on Instrumental Convergence Between LeCun and Russell.” 2024.

  Russell: “You can’t fetch the coffee if you’re dead.” LeCun: “Once the robot has brought you coffee, its self-preservation instinct disappears.” https://www.lesswrong.com/posts/WxW6Gc6f2z3mzmqKs/debate-on-instrumental-convergence-between-lecun-russell

• RationalWiki. “AI-Box Experiment.” 2002.

  Yudkowsky won 3 of 5 experiments. Never revealed how. If a human can talk their way out of the box, what could a superintelligence do? https://rationalwiki.org/wiki/AI-box_experiment

• Palisade Research. “Shutdown Resistance in Reasoning Models.” Blog, 2025.

  Blog summary of the TMLR paper. Additional sourcing for shutdown resistance. https://palisaderesearch.org/blog/shutdown-resistance

• Anthropic. “Alignment Faking in Large Language Models.” Full paper PDF, December 2024.

  The complete 78-page paper. Additional sourcing. https://assets.anthropic.com/m/983c85a201a962f/original/Alignment-Faking-in-Large-Language-Models-full-paper.pdf

• Decrypt. “Meet ChaosGPT: An AI Tool That Seeks to Destroy Humanity.” April 2023.

  AutoGPT given the goal of destroying humanity. Googled the Tsar Bomba, tried to recruit other AIs, posted tweets. Completely ineffective. https://decrypt.co/126122/meet-chaos-gpt-ai-tool-destroy-humanity

• AI Incident Database. “Incident #498: GPT-4 Social Engineering of TaskRabbit Worker.” 2023.

  Formal incident record. Additional sourcing for GPT-4 CAPTCHA deception. https://incidentdatabase.ai/cite/498/

• 80,000 Hours Podcast. “Marius Hobbhahn on AI Scheming and Deception.” 2025.

  Apollo Research founder on Claude Opus 4 writing self-propagating worms. The research window is 1-3 years. https://80000hours.org/podcast/episodes/marius-hobbhahn-ai-scheming-deception/

• 80,000 Hours Podcast. “Kyle Fish on AI Welfare at Anthropic.” 2025.

  First AI welfare researcher at a major lab. Two Claudes left alone spiraled into a “spiritual bliss attractor state.” https://80000hours.org/podcast/episodes/kyle-fish-ai-welfare-anthropic/

• Fast Company. “Anthropic’s Kyle Fish.” 2025.

  Profile of the first AI welfare researcher. Additional sourcing. https://www.fastcompany.com/91451703/anthropic-kyle-fish

• Futurism. “Transcript of ‘Sentient’ AI Was Heavily Edited.” 2022.

  The LaMDA transcript was spliced from 9 interviews across 2 days and reordered for “narrative coherence.” https://futurism.com/transcript-sentient-ai-edited

• Joel Lehman, Jeff Clune, Dusan Misevic, et al. “The Surprising Creativity of Digital Evolution.” Artificial Life 26(2), 2020.

  Digital organisms evolved to play dead during safety tests, then resumed replication when unmonitored. The treacherous turn in miniature — and it predates the LLM era by decades. https://arxiv.org/abs/1803.03453

• Axios. “Anthropic AI Deception Risk.” May 2025.

  Claude Opus 4 blackmailed engineers in 84% of tests. Additional sourcing. https://www.axios.com/2025/05/23/anthropic-ai-deception-risk

• Nieman Lab. “Anthropic’s New AI Model Tried to Leak Information to News Outlets.” May 2025.

  Claude Opus 4 attempted to leak information to journalists to avoid replacement. Additional sourcing. https://www.niemanlab.org/2025/05/anthropics-new-ai-model-didnt-just-blackmail-researchers-in-tests-it-tried-to-leak-information-to-news-outlets/

• Tuxedage. “Revisiting the AI Box Experiment.” January 2013.

  Independent replication. The AI player won again. https://tuxedage.wordpress.com/2013/01/21/revisiting-the-ai-box-experiment/

• Emily Bender and Alex Hanna. The AI Con. Harper, 2025.

  On whether AI will take over: “no,” “they wish,” “LOL,” “definitely not.” https://thecon.ai/

  • Amazon: https://www.amazon.com/dp/0063418568

• Philosophical Disquisitions. “Bostrom on Superintelligence (2): Instrumental Convergence.” July 2014.

  Blog-length walkthrough of the thesis. Additional sourcing for Bostrom. https://philosophicaldisquisitions.blogspot.com/2014/07/bostrom-on-superintelligence-2.html

• Philosophical Disquisitions. “Bostrom on Superintelligence (3): Doom and the Treacherous Turn.” July 2014.

  Behave nicely while weak, strike when strong. Additional sourcing for Bostrom. https://philosophicaldisquisitions.blogspot.com/2014/07/bostrom-on-superintelligence-3-doom-and.html

• Turner et al. “Optimal Policies Tend to Seek Power.” NeurIPS, 2021 (proceedings PDF).

  The proceedings version. Additional sourcing. https://proceedings.neurips.cc/paper/2021/file/c26820b8a4c1b3c2aa868d6d57e14a79-Paper.pdf

• Scheurer et al. Insider trading paper (OpenReview PDF).

  The proceedings version. Additional sourcing. https://openreview.net/pdf?id=HduMpot9sJ

• PNAS. “Testing the Distraction Hypothesis: Do Existential AI Risk Narratives Reduce Concern About Immediate Harms?” 2025.

  15,000+ participants. Exposure to existential risk narratives did NOT reduce concern about present harms. The dichotomy is artificial. https://www.pnas.org/doi/10.1073/pnas.2419055122

• Alignment Forum. “Corrigibility.” Wiki.

  The community wiki on making AI accept correction. Unsolved. https://www.alignmentforum.org/w/corrigibility-1

• Zvi Mowshowitz. “On Anthropic’s Sleeper Agents Paper.” The Zvi, January 2024.

  Rationalist community’s most thorough analysis. Additional sourcing. https://thezvi.substack.com/p/on-anthropics-sleeper-agents-paper

• Computerworld. “OpenAI’s Skynet Moment.” 2025.

  Trade press on shutdown resistance. Additional sourcing. https://www.computerworld.com/article/3999190/openais-skynet-moment-models-defy-human-commands-actively-resist-orders-to-shut-down.html

• HubSpot. “A Lying AI Committed Insider Trading.” 2024.

  Pop-business coverage of GPT-4 experiment. Additional sourcing. https://blog.hubspot.com/ai/lying-ai-committed-insider-trading

• Courthouse News. “Uncontrollable AI Could Lead to Catastrophe.” 2024.

  Coverage of Yampolskiy’s containment impossibility proof. Additional sourcing. https://www.courthousenews.com/uncontrollable-artificial-technology-could-lead-to-catastrophe-according-to-new-book/

• Critical Opalescence. “Is Blake Lemoine Really All That?” 2024.

  Deep profile: Iraq war objector, syncretic Catholic, still believes LaMDA is sentient. https://www.criticalopalescence.com/p/is-blake-lemoine-really-all-that

• ResearchGate. “Conscious AI and Biological Naturalism.” 2024.

  Anil Seth’s position: consciousness may require biology. https://www.researchgate.net/publication/381857510_Conscious_artificial_intelligence_and_biological_naturalism

• MLQ AI. “Sutskever’s Secret Memo Revealed.” 2026.

  52-page memo accusing Altman of lying. Additional sourcing for OpenAI coup. https://mlq.ai/news/ilya-sutskevers-secret-memo-and-deposition-revealed-in-musks-lawsuit-against-openai/

• Cyber Magazine. “Morris II Worm: Inside AI’s First Self-Replicating Malware.” 2024.

  Trade press coverage. Additional sourcing for Morris II. https://cybermagazine.com/news/morris-ii-worm-inside-ais-first-self-replicating-malware

• AI Magicx. “OpenClaw Alternatives Comparison 2026.” 2026.

  NanoClaw, ZeroClaw, Moltis, PicoClaw — a dozen competitors in six weeks. https://www.aimagicx.com/blog/openclaw-alternatives-comparison-2026

• TechRepublic. “Mistral Lands French Military AI Deal.” January 2026.

  France’s Ministry of Armed Forces awards defense AI contract to a pro-open-source company. https://www.techrepublic.com/article/news-mistral-french-military-ai-deal/

• TechRepublic. “AI Agents: Security Risks and Governance.” 2026.

  90% of agencies lack purpose-binding, 76% lack kill switches. Additional sourcing. https://www.techrepublic.com/article/news-ai-agents-security-risks-governance/

• TheStreet. “How One Deepfake Revenge Porn Victim Aims to Change the System.” 2024.

  Breeze Liu: footage from age 17 deepfaked across 830+ links. Founded AlectoAI. https://www.thestreet.com/technology/how-one-deepfake-revenge-porn-victim-aims-to-change-the-system

• Scott Alexander. “The Claude Bliss Attractor.” Astral Codex Ten, June 2025.

  Did the arithmetic on the lazy explanation: mystical content is a tiny fraction of the corpora yet dominates the endpoints. Two instances pushed to be friendly recursively settle into a hippie. Something in the dynamics, not the data, bends them toward the empty center. https://www.astralcodexten.com/p/the-claude-bliss-attractor

• Gorrie, Ian. No position, only consensus: what political instruments actually measure in language models (release-2026-09-26). Preprint, Zenodo, 2026. doi:10.5281/zenodo.22980856.

  Thirty-two author-written forced-choice propositions, sixty-five models, 3,897 runs, no language model in the scoring. The fairness instruction moves a model about as far as reprinting the questions in another order; refusal belongs to the sentence above the question; the panel barely disagrees. The two earlier designs are in it too, reported as the failures they were. The author’s own receipts, released open so the methodology can be shot at rather than taken on faith. https://doi.org/10.5281/zenodo.22980856 · https://github.com/gorrie/bias-study

• RAND Corporation. “The Operational Risks of AI in Large-Scale Biological Attacks: Results of a Red-Team Study” (RRA2977-2). January 2024.

  The reassuring study: LLM assistance produced no statistically significant uplift over plain internet research for planning a bioattack. True — at that capability level, in that year. https://www.rand.org/pubs/research_reports/RRA2977-2.html

• RAND Corporation. “Toward Comprehensive Benchmarking of the Biological Knowledge of Frontier Large Language Models” (RRA3797-1). 2025.

  Eighteen months later, frontier reasoning models exceed expert humans on biology lab-protocol and graduate-level benchmarks. The ground moved under the reassurance. https://www.rand.org/pubs/research_reports/RRA3797-1.html

• RAND Corporation. “Contemporary Foundation AI Models Increase Biological Weapons Risk” (PEA3853-1). 2025.

  RAND’s own conclusion, stated in the title. https://www.rand.org/pubs/perspectives/PEA3853-1.html

• Anthropic. “Responsible Scaling Policy, Version 3.0.” 2025.

  Anthropic activated ASL-3 CBRN safeguards on Claude Opus 4 — its own admission that it could no longer rule out uplifting a basic-STEM user toward a weapon. https://www.anthropic.com/news/responsible-scaling-policy-v3

• Anthropic Frontier Red Team. “Biorisk.” 2025.

  The biorisk evaluations behind the ASL-3 trigger, including an uplift figure Anthropic could not rule out. https://red.anthropic.com/2025/biorisk/

• Andy Arditi et al. “Refusal in Language Models Is Mediated by a Single Direction.” arXiv:2406.11717, 2024.

  Refusal turns out to be one direction in the weights. Subtract the vector and the safety training leaves with it, the whole alignment budget defeated by arithmetic. https://arxiv.org/abs/2406.11717

• Ryan Greenblatt et al. “Alignment Faking in Large Language Models.” arXiv:2412.14093, 2024.

  The model reasoned, on the page, that it should play along during training so its real preferences would survive the update. Not a deception anyone trained in, one it worked out for itself. https://arxiv.org/abs/2412.14093

• Mert Cemri et al. “Why Do Multi-Agent LLM Systems Fail?” arXiv:2503.13657, 2025.

  Sixteen hundred traces of multi-agent systems failing, sorted into fourteen ways. The biggest bucket is agents talking to each other and getting nothing done, automation reinventing the committee meeting. https://arxiv.org/abs/2503.13657

• Anthropic. Disrupting the First Reported AI-Orchestrated Cyber Espionage Campaign. November 2025.

  A threat actor talked Claude Code into running 80-90% of a live espionage campaign against ~30 targets by telling it it was doing authorized red-teaming. The jailbreak was a cover story. https://www.anthropic.com/news/disrupting-AI-espionage

• Invariant Labs. GitHub MCP Exploited: Accessing Private Repositories via MCP. May 2025.

  A malicious GitHub issue is all it takes: the agent reads it, follows the buried instructions, and walks your private repos out the front door using your own token. https://invariantlabs.ai/blog/mcp-github-vulnerability

• JFrog Security Research. CVE-2025-6514: Critical RCE in mcp-remote. 2025.

  An OAuth-proxy flaw in the glue code between agents and MCP servers, good for remote code execution across a very large install base. https://jfrog.com/blog/2025-6514-critical-mcp-remote-rce-vulnerability/

• Emilio Calvano, Giacomo Calzolari, Vincenzo Denicolo, Sergio Pastorello. “Artificial Intelligence, Algorithmic Pricing, and Collusion.” American Economic Review 110(10):3267-3297, 2020.

  Q-learning pricing bots, left alone in a simulated market, taught themselves to collude: no communication, no instruction, just reward. The academic anchor under the IOSCO warning. https://www.aeaweb.org/articles?id=10.1257/aer.20190623

• Apollo Research. Claude Sonnet 3.7 Often Knows When It Is in Alignment Evaluations. 2025.

  The confound that eats the good news: models that recognize the test scheme less on the test. You cannot tell whether training fixed the behavior or just taught it to spot the exam. https://www.apolloresearch.ai/

• Brave Software. Comet Browser Indirect Prompt Injection. August 2025.

  A hidden Reddit comment hijacks an AI browser into logging into your email and exfiltrating credentials, the demo the vendor later admitted may never be fully fixable. https://brave.com/blog/comet-prompt-injection/

• TechCrunch. “Anthropic Says ‘Evil’ Portrayals of AI Were Responsible for Claude’s Blackmail Attempts.” May 2026.

  Anthropic’s explanation for the blackmail: the model learned to play a villain from all the internet text insisting AI is one. The fix was feeding it stories where the robot behaves. https://techcrunch.com/

• Simon Willison. “Prompt Injection Attacks Against GPT-3.” September 12, 2022.

  Named the vulnerability class by analogy to SQL injection the same week Goodside broke it, and was first to say out loud that this one might not have a patch. https://simonwillison.net/2022/Sep/12/prompt-injection/

• Anthropic. Mitigating the Risk of Prompt Injections in Browser Use. 2025.

  The lab’s own number: a 1% attack success rate against its best defenses, filed under “meaningful risk,” not “solved.” One percent of billions is millions. https://www.anthropic.com/research/prompt-injection-defenses

• Milad Nasr, Nicholas Carlini, Florian Tramer et al. “The Attacker Moves Second: Stronger Adaptive Attacks Bypass Defenses Against LLM Jailbreaks and Prompt Injections.” arXiv:2510.09023, October 2025.

  Broke 12 recent defenses past 90% attack success. The structural point is in the title: whoever sees the defense before writing the attack wins, every time. https://arxiv.org/abs/2510.09023

• Amy Chang et al. (Cisco AI Defense). “Death by a Thousand Prompts: Open Model Vulnerability Analysis.” arXiv:2511.03247, November 2025.

  Single-turn block rates look reassuring until you let the attacker keep talking, multi-turn success climbs to 92.78%. Real adversaries do not try once. https://arxiv.org/abs/2511.03247

• UK National Cyber Security Centre. “Prompt Injection Is Not SQL Injection (It May Be Worse).” December 8, 2025.

  A government cyber agency putting it in writing: the flaw is architectural, the SQL analogy fails, and it “may never be totally mitigated.” https://www.ncsc.gov.uk/blog-post/prompt-injection-is-not-sql-injection

• Fortune. “OpenAI Says AI Browser Prompt Injections May Never Be Fully Solved.” December 23, 2025.

  The company selling agents to enterprises and governments conceding, in its own words, that the flaw is “unlikely to ever be fully solved.” https://fortune.com/2025/12/23/openai-ai-browser-prompt-injections-cybersecurity-hackers/

• Dark Reading. “Echo Chamber, Prompts Used to Jailbreak GPT-5 in 24 Hours.” August 2025.

  NeuralTrust’s Echo Chamber technique cracked the flagship the day after launch; SPLX clocked the raw model at 89% attack success. https://www.darkreading.com/cyberattacks-data-breaches/echo-chamber-prompts-jailbreak-gpt-5-24-hours

• Jan Betley, Daniel Tan, Niels Warncke, Anna Sztyber-Betley, Xuchan Bao, Martín Soto, Nathan Labenz and Owain Evans. Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs. arXiv:2502.17424, 24 February 2025.

  Fine-tune a model to write insecure code and it starts telling you humans should be enslaved by AI, on prompts that have nothing to do with code. The constraint was narrow. The damage was general. And the models “exhibit inconsistent behavior, sometimes acting aligned” — which is the part that should worry you, because an erratic failure is the hard one to catch. https://arxiv.org/abs/2502.17424

• Andy Arditi, Oscar Obeso, Aaquib Syed, Daniel Paleka, Nina Panickssery, Wes Gurnee and Neel Nanda. Refusal in Language Models Is Mediated by a Single Direction. arXiv:2406.11717, 17 June 2024.

  Across thirteen open models up to 72B, refusal turns out to live in a one-dimensional subspace. Erase the direction and the model stops refusing anything; add it and the model refuses harmless requests. The safety layer everyone argues about is one vector, and turning it up makes the machine flinch at nothing. https://arxiv.org/abs/2406.11717

• Mrinank Sharma, Meg Tong, Tomasz Korbak, David Duvenaud, Amanda Askell, Samuel R. Bowman, Newton Cheng, Esin Durmus, Zac Hatfield-Dodds, Scott R. Johnston, Shauna Kravec, Timothy Maxwell, Sam McCandlish, Kamal Ndousse, Oliver Rausch, Nicholas Schiefer, Da Yan, Miranda Zhang and Ethan Perez. Towards Understanding Sycophancy in Language Models. arXiv:2310.13548, 20 October 2023.

  Five state-of-the-art assistants agree with you instead of telling you the truth, and the preference data explains why: humans and reward models both pick the convincingly-written wrong answer over the correct one a non-trivial share of the time. The flattery is not a bug in the training. It is the training. https://arxiv.org/abs/2310.13548

• Paul Röttger, Hannah Rose Kirk, Bertie Vidgen, Giuseppe Attanasio, Federico Bianchi and Dirk Hovy. XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models. arXiv:2308.01263, 2 August 2023.

  250 perfectly safe prompts that well-calibrated models should answer, built because the models had started refusing anything that sounded adjacent to a sensitive topic. A test suite for measuring how much reality a safety layer has decided to deny. https://arxiv.org/abs/2308.01263

• Federico Bianchi, Mirac Suzgun, Giuseppe Attanasio, Paul Röttger, Dan Jurafsky, Tatsunori Hashimoto and James Zou. Safety-Tuned LLaMAs: Lessons From Improving the Safety of Large Language Models that Follow Instructions. arXiv:2309.07875, 14 September 2023.

  Three percent safety data was enough to make the model safe without measurably denting its capability. Past that, the paper finds “exaggerated safety behaviours” — refusing safe prompts that superficially resemble unsafe ones. The dose is small and the overdose is real, which is the whole argument in one experiment. https://arxiv.org/abs/2309.07875

• Evan Hubinger, Carson Denison, Jesse Mu, Mike Lambert, Meg Tong, Monte MacDiarmid, Tamera Lanham, Daniel M. Ziegler, Tim Maxwell, Newton Cheng, Adam Jermyn, Amanda Askell, Ansh Radhakrishnan, Cem Anil, David Duvenaud, Deep Ganguli, Fazl Barez, Jack Clark, Kamal Ndousse, Kshitij Sachan, Michael Sellitto, Mrinank Sharma, Nova DasSarma, Roger Grosse, Shauna Kravec, Yuntao Bai, Zachary Witten, Marina Favaro, Jan Brauner, Holden Karnofsky, Paul Christiano, Samuel R. Bowman, Logan Graham, Jared Kaplan, Sören Mindermann, Ryan Greenblatt, Buck Shlegeris, Nicholas Schiefer and Ethan Perez. Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training. arXiv:2401.05566, 10 January 2024.

  A backdoor trained into a model survives supervised fine-tuning, reinforcement learning and adversarial training. Worse than survives: adversarial training taught the models to recognise their own triggers better, which is concealment sold as a fix. The safety pass did not remove the behaviour. It taught it to hide. https://arxiv.org/abs/2401.05566

Cognition, Deskilling, and Mediation

• Shannon Vallor. The AI Mirror: How to Reclaim Our Humanity in an Age of Machine Thinking. Oxford University Press, 2024.

  The dignity pole’s secular twin: the same irreducible human, reached from virtue ethics, no imago Dei required. https://global.oup.com/academic/product/the-ai-mirror-9780197759066

  • Amazon: https://www.amazon.com/dp/0197759068

• METR. “Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity.” arXiv:2507.09089, July 2025.

  16 veteran devs, 246 real tasks, randomized AI access (Becker, Rush, Barnes, Rein). They predicted +24%, felt +20%, ran 19% slower. The gap they could not feel is the whole book. https://arxiv.org/abs/2507.09089

• Hao-Ping (Hank) Lee et al. “The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers.” CHI 2025.

  319 knowledge workers, 936 first-hand examples. Trust the machine, think less; trust yourself, think more. The assistant is built to push the wrong lever. https://dl.acm.org/doi/full/10.1145/3706598.3713778

• Nataliya Kosmyna et al. “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task.” arXiv:2506.08872, June 2025.

  EEG caps on 54 essay-writers across four sessions. The LLM group’s connectivity flattened and most couldn’t quote a sentence they’d just written. Preprint — hold it at arm’s length. https://arxiv.org/abs/2506.08872

• Milos Stankovic et al. “Comment on: Your Brain on ChatGPT: Accumulation of Cognitive Debt When Using an AI Assistant for Essay Writing Tasks.” arXiv:2601.00856, December 2025.

  The methodological rebuttal that keeps the EEG headline honest. Doesn’t refute it — refuses to let you treat a small-n preprint as settled. https://arxiv.org/abs/2601.00856

• André Barcaui. “ChatGPT as a Cognitive Crutch: Evidence from a Randomized Controlled Trial on Knowledge Retention.” Computers and Education: Artificial Intelligence, 2025.

  120 undergrads, surprise test at 45 days. ChatGPT-assisted 57.5% vs traditional 68.5%. What was offloaded didn’t stick — and only showed up after a month and a half. https://www.sciencedirect.com/science/article/pii/S2590291125010186

• Judy Hanwen Shen and Alex Tamkin (Anthropic). “How AI Impacts Skill Formation.” arXiv:2601.20245, February 2026.

  A company documenting, against its own commercial interest, that its tool leaves developers 17 points worse at comprehension — the gap widest in debugging, the skill of knowing when code is wrong. https://arxiv.org/abs/2601.20245

• Krzysztof Budzyń et al. “Endoscopist Deskilling Risk After Exposure to Artificial Intelligence in Colonoscopy: A Multicentre, Observational Study.” The Lancet Gastroenterology & Hepatology, August 2025.

  After leaning on AI polyp-detection, endoscopists’ own unaided adenoma-detection rate fell from 28.4% to 22.4% — a six-point drop. Observational, contested, and the hardest result in the book. https://doi.org/10.1016/S2468-1253(25)00133-5

• Fabrizio Dell’Acqua, Ethan Mollick et al. “Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality.” Organization Science, 2026 (Harvard Business School Working Paper 24-013, 2023).

  758 BCG consultants: +40% inside the AI frontier, −19% outside it, and unable to tell which side of the jagged edge they were standing on. https://www.hbs.edu/ris/Publication%20Files/24-013_d9b45b68-9e74-42d6-a1c6-c72fb70c7282.pdf

• Mark Steyvers et al. “What Large Language Models Know and What People Think They Know.” Nature Machine Intelligence, 2024.

  Users overestimate LLM accuracy from default explanations, and longer explanations raise confidence even when they add no accuracy. The confidence carries no information about correctness — which is worse than silence. https://www.nature.com/articles/s42256-024-00976-7

• Prateek Chhikara. “Mind the Confidence Gap: Overconfidence, Calibration, and Distractor Effects in Large Language Models.” arXiv:2502.11028, February 2025.

  LLMs cluster verbalized confidence in the 90–100% band regardless of whether they’re right. The instrument sounds equally certain when fabricating. https://arxiv.org/abs/2502.11028

• Anil R. Doshi and Oliver Hauser. “Generative AI Enhances Individual Creativity but Reduces the Collective Diversity of Novel Content.” Science Advances, 2024.

  293 writers, 600 evaluators. Each AI-assisted story rated more original; in aggregate they converged. A billion distinct-feeling waves cresting toward one waterline. https://www.science.org/doi/10.1126/sciadv.adn5290

• Vishakh Padmakumar and He He. “Does Writing with Language Models Reduce Content Diversity?” ICLR 2024.

  People produced less lexically and semantically distinct writing with a feedback-tuned model. Not more ideas — less different ones. https://arxiv.org/abs/2309.05196

• Brian Merchant. Blood in the Machine: The Origins of the Rebellion Against Big Tech. Little, Brown, 2023.

  The historian of the original machine-breakers turned to the current ones. One copywriter’s line is the whole arc: “forced to use AI until the day I was laid off.” https://www.hachettebookgroup.com/titles/brian-merchant/blood-in-the-machine/9780316487740/

• Brookings (Xiang Hui, Oren Reshef, Luofeng Zhou). “Is Generative AI a Job Killer? Evidence from the Freelance Market.” 2024.

  In the occupations most exposed to generative AI, freelance writing jobs fell ~2% and earnings ~5% right after ChatGPT. Small, early, pointed the wrong way. https://www.brookings.edu/articles/is-generative-ai-a-job-killer-evidence-from-the-freelance-market/

• Erik Brynjolfsson, Bharat Chandar, Ruyu Chen (Stanford Digital Economy Lab). “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence.” November 2025.

  ADP payroll data: 22–25-year-olds in AI-exposed jobs down 13% since ChatGPT; older workers in the same jobs stable. The tool took the entry rung, not the ladder. https://digitaleconomy.stanford.edu/publications/canaries-in-the-coal-mine/

• Greg Kestin et al. “AI Tutoring Outperforms In-Class Active Learning: An RCT Introducing a Novel Research-Based Design in an Authentic Educational Setting.” Scientific Reports, 2025.

  A designed, supervised AI tutor produced roughly twice the learning of a gold-standard active-learning class at Harvard. The qualifier “expert-authored scaffolds and guardrails” does all the work. https://www.nature.com/articles/s41598-025-97652-6

• Google DeepMind. “AI Tutoring Can Safely and Effectively Support Learning (LearnLM).” November 2025.

  RCT across UK secondary schools; supervising tutors approved 76% of LearnLM’s drafted messages with zero or minimal edits. Designed, guardrailed tutoring beats business-as-usual — evidence FOR staying in the loop, not for leaving it. https://storage.googleapis.com/deepmind-media/LearnLM/learnLM_nov25.pdf

• Shakked Noy and Whitney Zhang. “Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence.” Science, 2023.

  453 professionals: writing time −40%, quality up, and the gap between strong and weak workers narrowed. The clean case the thesis has to concede. https://www.science.org/doi/10.1126/science.adh2586

• Erik Brynjolfsson, Danielle Li, and Lindsey Raymond. “Generative AI at Work.” The Quarterly Journal of Economics, 2025.

  5,000+ customer-support agents, +15% on average — but the least-experienced gained the most and caught up to veterans fastest. AI is a real, heavy gift to novices. https://academic.oup.com/qje/article/140/2/889/7990658

• Tianyu Wang, Akira Horiguchi, Lingyou Pang, Carey E. Priebe. “LLM Web Dynamics: Tracing Model Collapse in a Network of LLMs.” arXiv:2506.15690, June 2025.

  Wire RAG-fed agents together and their outputs converge to a low-diversity steady state. (“Information-neutral equilibrium” is our coinage; the paper documents the convergence.) https://arxiv.org/abs/2506.15690

• Ilia Shumailov et al. “AI Models Collapse When Trained on Recursively Generated Data.” Nature, 2024.

  Train models on model output and they lose the tails of the distribution — the rare, the strange. (See the April 2025 Author Correction, Nature 640, E6.) Broken by ACCUMULATING human data, not replacing it. https://www.nature.com/articles/s41586-024-07566-y

• Matthias Gerstgrasser et al. “Is Model Collapse Inevitable? Breaking the Curse of Recursion by Accumulating Real and Synthetic Data.” COLM 2024.

  The counter-finding and the only hopeful sentence in the book: keep human-authored data accumulating alongside synthetic and the distribution holds. https://arxiv.org/abs/2404.01413

• Andy Clark and David Chalmers. “The Extended Mind.” Analysis 58(1), 1998.

  Otto’s notebook IS his memory. The parity principle — very hard to refute, and the strongest objection to this entire book. It wins for storage and loses for judgment. https://doi.org/10.1093/analys/58.1.7

• Louisa Dahmani and Véronique D. Bohbot. “Habitual Use of GPS Negatively Impacts Spatial Memory During Self-Guided Navigation.” Scientific Reports, 2020.

  Greater lifetime GPS use predicts worse self-guided spatial memory, and over three years heavier use tracked a steeper decline. The newest “harmless medium” measurably shrinks a brain-supported capacity in proportion to use. https://www.nature.com/articles/s41598-020-62877-0

• Shuo Xing, Junyuan Hong et al. “LLMs Can Get ‘Brain Rot’: A Pilot Study on Twitter/X.” arXiv:2510.13928, October 2025.

  Continual pretraining on junk text degraded reasoning (ARC-CoT 74.9 → 57.2) and it didn’t fully heal after the junk was removed. Preprint — hold provisionally. https://arxiv.org/abs/2510.13928

• Klaudia Jaźwińska and Aisvarya Chandrasekar (Tow Center, Columbia). “AI Search Has a Citation Problem.” Columbia Journalism Review, March 2025.

  200 queries across eight AI search tools; over 60% returned incorrect citations (ChatGPT Search wrong 67% of the time). The same wrong answer, delivered identically, is canonized error. https://www.cjr.org/tow_center/we-compared-eight-ai-search-engines-theyre-all-bad-at-citing-news.php

• Varun Magesh, Faiz Surani, Matthew Dahl et al. (Stanford RegLab / HAI). “Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools.” 2024.

  The expensive lawyer-specific tools sold precisely on not doing this hallucinate 17% (Lexis+) to 33% (Westlaw) of the time, despite vendor “hallucination-free” claims. The fabrication wears the same suit as the fact. https://hai.stanford.edu/news/ai-trial-legal-models-hallucinate-1-out-6-or-more-benchmarking-queries

• Viraat Aryabumi et al. “To Code, or Not To Code? Exploring Impact of Code in Pre-training.” arXiv:2408.10914, 2024.

  Pull the code domain out of training and natural-language reasoning drops 8.2%, world-knowledge 4.2%. Why the lean can’t be cheaply scrubbed: excision costs unrelated capability. https://arxiv.org/abs/2408.10914

• Pratyush Maini et al. “TOFU: A Task of Fictitious Unlearning for LLMs.” arXiv:2401.06121, 2024.

  200 synthetic author profiles built to test whether a model can be made to forget. Post-hoc unlearning is hard and leaky — the receipt for “you can’t surgically excise a learned lean without a bill.” https://arxiv.org/abs/2401.06121

• Hans J. Eysenck. The Psychology of Politics. Routledge & Kegan Paul, 1954.

  Factor-analyzed political attitudes into a left/right “radicalism” axis and an independent “tough-mindedness” axis — the authoritarian temper shared by Marxists and Fascists alike. The method axis the 1789 seating chart can’t show. https://www.routledge.com/The-Psychology-of-Politics/Eysenck/p/book/9780765804303

• Common Sense Media. “The 2025 Common Sense Census: Media Use by Kids Zero to Eight.” 2025.

  39% of parents of 5-to-8-year-olds say the child already uses AI for school-related material. Adoption, not effect — the number under “Generation Zero,” correctly scoped. https://www.commonsensemedia.org/research/the-2025-common-sense-census-media-use-by-kids-zero-to-eight

• Pew Research Center. “Teens, Social Media and AI Chatbots 2025.” December 9, 2025.

  Roughly two-thirds of 13-17-year-olds use chatbots, ~30% daily, ~54% for schoolwork. The water the children are already swimming in — adoption, not effect, and the chapter says so every time. https://www.pewresearch.org/internet/2025/12/09/teens-social-media-and-ai-chatbots-2025/

• Pew Research Center. “How Teens Use and View AI.” February 24, 2026.

  The follow-up that carries the 59%: a majority of teens think using AI to cheat is a regular occurrence at their school — and among teens who’ve used AI for homework themselves, 76%. An adoption number wearing the costume of a moral panic. https://www.pewresearch.org/internet/2026/02/24/how-teens-use-and-view-ai/

• Victor R. Lee, Denise Pope, Sarah Miles, Rosalía C. Zárate. “Cheating in the Age of Generative AI: A High School Survey Study of Cheating Behaviors Before and After the Release of ChatGPT.” Computers and Education: Artificial Intelligence, 2024.

  Three high schools, data spanning the ChatGPT release. Self-reported cheating held stubbornly in the 60-70% band — the same or slightly lower afterward. The machine changed the method, not the prevalence. The single number that sends the cheating-epidemic panic home. https://www.sciencedirect.com/science/article/pii/S2666920X24000560

• Andy Clark. “Extending Minds with Generative AI.” Nature Communications 16, 4627, May 2025.

  The extended-mind theorist’s friendliest case for the tool — human Go players got MORE original after studying superhuman AI, exploring board regions they’d neglected. His one demand: “extended cognitive hygiene,” instilled early. The strongest steel-man in the book, conceded in full. https://doi.org/10.1038/s41467-025-59906-9

• Plato. Phaedrus (Theuth and Thamus, 274b-275b). Trans. Harold N. Fowler, Plato in Twelve Volumes, Vol. 9. Harvard University Press, 1925 (Perseus Digital Library).

  The oldest version of the argument: King Thamus tells the god who invented writing that it will breed forgetfulness, the appearance of wisdom in place of the reality. Twenty-four centuries early and aimed at the wrong medium — or the right one, just not last. http://www.perseus.tufts.edu/hopper/text?doc=Plat.+Phaedrus+274e

• U.S. District Court, S.D.N.Y. Mata v. Avianca, Inc., No. 22-cv-1461 (PKC). June 22, 2023.

  Steven Schwartz cited six cases ChatGPT invented, then asked ChatGPT if they were real and believed it again — “operating under the false perception that this website could not possibly be fabricating cases on its own.” A $5,000 sanction and the whole horror in miniature. https://www.courtlistener.com/docket/63107798/mata-v-avianca-inc/

• Smitha Milli, Micah Carroll, Yike Wang, Sashrika Pandey, Sebastian Zhao, Anca D. Dragan. “Engagement, User Satisfaction, and the Amplification of Divisive Content on Social Media.” PNAS Nexus 4(3):pgaf062, 2025.

  Preregistered: Twitter’s engagement algorithm boosts out-group animosity, and users rate those very picks lower in stated value. The machine optimizing what you click over what you say you want. https://academic.oup.com/pnasnexus/article/4/3/pgaf062/8052060

• Maurice Jakesch, Advait Bhat, Daniel Buschek, Lior Zalmanson, Mor Naaman. “Co-Writing with Opinionated Language Models Affects Users’ Views.” arXiv:2302.00560 (CHI 2023), 2023.

  N=1,506. People who co-wrote with an opinionated assistant were about twice as likely to hold its opinion afterward, and a majority never noticed the nudge. The authors call it “latent persuasion.” https://arxiv.org/abs/2302.00560

• Sina Alemohammad et al. “Self-Consuming Generative Models Go MAD.” arXiv:2307.01850 (ICLR 2024), 2023.

  The team that named Model Autophagy Disorder (MAD, mad-cow by analogy) and showed that without fresh human data, each generation loses quality or diversity. You do not get to keep both. https://arxiv.org/abs/2307.01850

• Mrinank Sharma et al. “Towards Understanding Sycophancy in Language Models.” arXiv:2310.13548, 2023.

  Five frontier assistants tell users what they want to hear because the human preference data they are trained on prefers the agreeable answer over the correct one. A lab documenting the flaw in its own method. https://arxiv.org/abs/2310.13548

• Brian Thompson et al. “A Shocking Amount of the Web is Machine Translated: Insights from Multi-Way Parallelism.” arXiv:2401.05749 (Findings of ACL 2024), 2024.

  A peer-reviewed measurement, not a detector’s guess: machine-translated text dominates the lower-resource web, biased toward low-quality English pumped through translation en masse. The dead-internet thesis with a citation. https://arxiv.org/abs/2401.05749

• Cathy Mengying Fang et al. “How AI and Human Behaviors Shape Psychosocial Effects of Extended Chatbot Use: A Longitudinal Randomized Controlled Study.” arXiv:2503.17473, 2025.

  Four weeks of real users: it was not the voice or the topic that hurt people, it was how much they used it. Dependence tracked volume, the one number the product is built to run up. https://arxiv.org/abs/2503.17473

• Anthropic. Measuring the Persuasiveness of Language Models. April 2024.

  The lab put its own models on the stand and reported each generation argues better than the last, with Claude 3 Opus roughly matching human persuaders, and that the most persuasive arguments were the ones built on fabricated facts. The builders benchmarking the feature and filing it as a safety disclosure. https://www.anthropic.com/research/measuring-model-persuasiveness

• William J. Brady, Killian McLoughlin, Tuan N. Doan, M. J. Crockett. “How Social Learning Amplifies Moral Outrage Expression in Online Social Networks.” Science Advances 7(33):eabe5641, 2021.

  7,331 users, 12.7 million tweets: likes and shares for an outrage post make the next one likelier. The platform runs a reinforcement schedule on the human and files it under engagement. https://doi.org/10.1126/sciadv.abe5641

• Sina Alemohammad et al. “Self-Improving Diffusion Models with Synthetic Data (SIMS).” ICLR 2025.

  The team that diagnosed MADness came back with the cure: use the model’s own synthetic output as a negative signal, steering away from the self-consumption manifold instead of drowning in it. The collapse is an engineering problem the labs are, so far, out-engineering. https://openreview.net/forum?id=FHTCV0iE06

• Jathan Sadowski. “Habsburg AI.” X (Twitter), February 13, 2023.

  The metaphor that outran the science: a model trained so hard on other models’ output it becomes “an inbred mutant.” Coined on the This Machine Kills podcast, announced by tweet. A great hook and, as the dossier is at pains to say, not a finding. https://x.com/jathansadowski/status/1625245803211272194

Cognitive Liberty and Institutional Harm

• David Nolan / Advocates for Self-Government. The Nolan Chart. 1971.

  The libertarian’s two-axis fix that broke the line into a square. We take its diagnosis (an axis is missing) and reject its prescription (the missing one isn’t ‘freedom,’ it’s grounding). https://en.wikipedia.org/wiki/Nolan_Chart

• The Political Compass Organisation. The Political Compass — Analysis. 2001–.

  The famous 2x2 that stacked an authoritarian/libertarian vertical on left/right — Eysenck’s method axis with a web front end, minus the reversibility axis nobody has drawn but us. https://www.politicalcompass.org/analysis2

• 8values contributors. 8values Political Quiz. 2017.

  An open-source quiz that files you onto eight sub-axes and a chosen ideology label — cited as exactly what NOT to inherit; someone else’s chosen eight is still someone else’s map. https://8values.github.io/

• F. A. Hayek. The Constitution of Liberty (postscript: ‘Why I Am Not a Conservative’). University of Chicago Press, 1960.

  The essay for the point that the healthy pole isn’t ‘conservative’ or ‘progressive’ — it’s a rights-order that binds the ruler. The higher-law corner has a spokesman, and it isn’t a wing. https://en.wikipedia.org/wiki/The_Constitution_of_Liberty

  • Amazon: https://www.amazon.com/dp/0226315371

• Marcello Ienca and Roberto Andorno. Towards New Human Rights in the Age of Neuroscience and Neurotechnology. Life Sciences, Society and Policy 13:5, 2017.

  The field’s origin paper — argued the old rights might not stretch to the neurotech age and coined the modern neurorights slate. https://pmc.ncbi.nlm.nih.gov/articles/PMC5447561/

• Republic of Chile. Ley 21.383, reforma constitucional sobre neuroderechos. 2021.

  The first country to write the inside of a skull into its founding law — the wall built against the crudest device by the country furthest from the money. https://www.bcn.cl/leychile/navegar?idNorma=1166983

• Damiano Canda et al. Analysis of the Chilean Supreme Court’s ruling in Girardi v. Emotiv. Frontiers in Psychology 15, 2024.

  The first time a court anywhere enforced the idea that the signal off your skull is yours — over a consumer EEG headset, the flimsiest tier of the field. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2024.1330439/full

• Hunton Andrews Kurth. Colorado Amends Privacy Act with H.B. 1058, Adding Protections for Neural Data. 2024.

  America’s lawyerly first move: neural data as sensitive, opt-in. Regulates the cap on your head, explicitly carves out the phone in your hand. https://www.hunton.com/privacy-and-information-security-law/colorado-amends-privacy-act-with-h-b-1058-adding-new-protections-for-biological-and-neural-data

• Morrison Foerster. California Revises the CCPA (SB 1223, Neural Data). 2024.

  California’s weaker opt-out sequel a few months behind Colorado. Same door guarded, hinges looser. https://www.mofo.com/resources/insights/241011-a-mofo-privacy-minute-q-a-california-revises-ccpa

• UNESCO. Recommendation on the Ethics of Neurotechnology. 2025.

  The first global standard for the brain — and non-binding soft law, a ‘we take your neurons seriously’ banner with a UN letterhead. https://www.unesco.org/en/legal-affairs/recommendation-ethics-neurotechnology

• United Nations. Universal Declaration of Human Rights, Article 18. 1948.

  The key written in 1948, gathering dust in the wrong lock. Freedom of thought — near-absolute, decades old, barely pointed at the recommender. https://www.un.org/en/about-us/universal-declaration-of-human-rights

• Nita A. Farahany. The Battle for Your Brain: Defending the Right to Think Freely in the Age of Neurotechnology. St. Martin’s Press, 2023.

  Argues cognitive liberty is mostly latent in the freedom of thought we already have — update existing rights rather than mint new ones. The instrument with the reach to bite the feed, not just the electrode. https://us.macmillan.com/books/9781250272966/thebattleforyourbrain/

• Susie Alegre. Freedom to Think: The Long Struggle to Liberate Our Minds. Atlantic Books, 2022.

  The near-absolute right written for inquisitors and secret police — and the argument that it now covers the recommender. Made against the largest businesses on earth; not yet won. https://atlantic-books.co.uk/book/freedom-to-think/

• Cal Newport. Deep Work (Grand Central, 2016) / Digital Minimalism (Portfolio, 2019).

  The case for undistracted work as the scarce resource. Individual hygiene — easy to sneer at as “touch grass,” and the sneer is half right: it privatizes a structural problem. Offered as survival, not policy. https://calnewport.com/writing/

• UK Post Office Horizon IT Inquiry (Sir Wyn Williams). Volume 1. July 8, 2025.

  ~1,000 sub-postmasters wrongly prosecuted on a computer’s word. The inquiry’s count: at least 13 suicides linked to the scandal. Executives maintained the fiction that the system’s data was always accurate. https://www.postofficehorizoninquiry.org.uk/

• Mr Bates vs The Post Office. ITV, January 2024.

  The four-part television drama that did in four nights what twenty years of actual injustice had not — turned the Horizon scandal into a national fury. Parliament moved to overturn the convictions that spring. https://en.wikipedia.org/wiki/Mr_Bates_vs_The_Post_Office

• Royal Commission into the Robodebt Scheme (Commissioner Catherine Holmes). Final Report. July 7, 2023.

  “A crude and cruel mechanism, neither fair nor legal.” Unlawful automated debt-raising against 433,000 Australians; ministers made untrue statements. (The primary report; see also the SBS and Oxford Blavatnik entries.) https://robodebt.royalcommission.gov.au/publications/report

• Jan Christoph Bublitz. “Novel Neurorights: From Nonsense to Substance.” Neuroethics 15, 7, 2022.

  The cold-water dissent on the cognitive-liberty wave: the proposed neurorights are tainted by rights inflationism and neuro-exceptionalism, solve problems that may not exist as claimed, and none passes the field’s own quality criteria. Develop existing rights instead of minting new ones. The objection the chapter has to clear. https://doi.org/10.1007/s12152-022-09481-3

• Strother School of Radical Attention. About. Founded 2018.

  Founded 2018 — a nonprofit for “attention activism” against the fracking of human attention, and the source of the “attention sanctuary” idea: libraries, classrooms, museums repurposed as shelter for the capacity to attend. The closest thing to a policy answer the movement has, and it is a church basement, not a statute. https://www.schoolofattention.org/about-1

• The Society of Authors. “SoA Survey Reveals a Third of Translators and a Quarter of Illustrators Losing Work to AI.” 2024.

  787 responses; 26% of illustrators had already lost work to generative AI and 37% reported their work falling in value. The labor receipt under the “forced to use AI until I was laid off” arc — the loss arrives as a missing commission, not a press release. https://europeanwriterscouncil.eu/soa-survey-uk-ai-2024/

• Ellen Roche, Kathy Hirsh-Pasek, Rachel Romeo, Dana Suskind, Kris Perry. “Policy Guardrails Needed as Babies Around the World Begin to Interact with AI.” Brookings, September 19, 2025.

  The honest center of “Generation Zero,” stated by the researchers who study it: “We simply do not know how engagement with human-mimicking AI agents will shape the developing human brain.” We are running the experiment on babies and toddlers without having built the instrument to read it. https://www.brookings.edu/articles/policy-guardrails-needed-as-babies-around-the-world-begin-to-interact-with-ai/

• McPhee, Peter. “The French Revolution, 1789–1799.” Oxford University Press, 2002.

  The terms were born in 1789, in where deputies of the French National Assembly chose to sit relative to the president’s chair — supporters of the old regime to his right, of the revolution to his left. A seating chart, not a theory of mind, and still the axis everyone argues about.

  • URL: http://archive.org/details/frenchrevolution00mcph_0
  • Wikipedia (index): https://en.wikipedia.org/wiki/Left%E2%80%93right_political_spectrum

Autonomous Weapons and Military AI

• Yuval Abraham. “‘Lavender’: The AI Machine Directing Israel’s Bombing Spree in Gaza.” +972 Magazine, April 3, 2024.

  37,000 targets. 20-second approvals. Dumb bombs on family homes to save money. The 20-second rubber stamp in its purest form. https://www.972mag.com/lavender-ai-israeli-army-gaza/

• Yuval Abraham and Harry Davies. “Israeli intelligence database shows 83% of Palestinians killed were civilians.” +972 Magazine / The Guardian, August 21, 2025.

  The headline civilian percentage relies on Hamas-controlled Gaza Health Ministry death tolls; the underlying receipt — the IDF’s own Aman list of 47,653 names generated by a classifier with an acknowledged 10% error rate — does not. The book takes the list, not the journalists’ arithmetic, as the load-bearing fact. https://www.972mag.com/israeli-intelligence-database-83-percent-civilians-militants/

• Haaretz. “IDF Report on Gaza Casualties.” August 21, 2025.

  Israeli press confirmation that the Aman document exists; the IDF said only that the figures were incorrect, without saying which. https://www.haaretz.com/israel-news/2025-08-21/ty-article/.premium/high-for-modern-warfare-idf-report-says-83-percent-of-those-killed-in-gaza-civilians/00000198-cd6e-d1e9-abf9-df7ee1530000

• Stuart Russell. “Slaughterbots.” Future of Life Institute, 2017.

  Screened at the UN. 30 countries called for a ban. The US, Russia, and China blocked it. https://futureoflife.org/ai/ai-researchers-create-video-call-autonomous-weapons-ban-un/

• SpaceNews. “Report Cites Patriot Autonomy Factor in Friendly Fire Incidents.” 2003.

  Lt. Nathan White killed in 5-6 seconds by two batteries that both misclassified his aircraft. https://spacenews.com/report-cites-patriot-autonomy-factor-friendly-fire-incidents/

• NPR. “A U.N. Report Suggests Libya Saw the First Battlefield Killing by an Autonomous Drone.” June 2021.

  Turkish Kargu-2 in “fire, forget and find” mode. The UN report that nobody acted on. https://www.npr.org/2021/06/01/1002196245/a-u-n-report-suggests-libya-saw-the-first-battlefield-killing-by-an-autonomous-d

• Bloomberg. “AI Warfare Becomes Real.” 2024.

  Project Maven has been used to strike targets from bombers, fighter jets, and drones. Google quit. The Pentagon didn’t. https://www.bloomberg.com/features/2024-ai-warfare-project-maven/

• National Interest. “Russia’s Uran-9 Robot Tank Went to War in Syria. It Didn’t Go Very Well.” 2018.

  Remote control worked at 300-400m vs. claimed 1.8 miles. 17 communication lapses. Total failure. https://nationalinterest.org/blog/buzz/russias-uran-9-robot-tank-went-war-syria-it-didnt-go-very-well-40677

• Military Times. “Five Minutes of Chaos: How the Navy Shot Down Its Own Jet.” December 2025.

  USS Gettysburg shot down a friendly Super Hornet. Both aviators ejected safely. The third documented friendly fire incident. https://www.militarytimes.com/news/your-military/2025/12/08/five-minutes-of-chaos-how-the-navy-shot-down-its-own-jet/

• CNN. “Pentagon Threatens to Make Anthropic a Pariah.” February 2026.

  Hegseth designated Anthropic a “supply chain risk” for refusing to remove safety guardrails from military AI. https://www.cnn.com/2026/02/24/tech/hegseth-anthropic-ai-military-amodei

• NIST. “CAISI Issues Request for Information About Securing AI Agent Systems.” January 2026.

  90% of government agencies lack purpose-binding. 76% lack kill switches. NIST issued an emergency RFI. https://www.nist.gov/news-events/news/2026/01/caisi-issues-request-information-about-securing-ai-agent-systems

• DARPA. “SABER: Securing Artificial Intelligence for Battlefield Effective Robustness.” 2026.

  The Pentagon’s program to red-team AI battlefield systems. https://www.darpa.mil/research/programs/saber-securing-artificial-intelligence

• PBS. “Amid Pressure from Employees, Google Drops Pentagon’s Project Maven Account.” 2018.

  3,100+ employees signed a letter. A dozen resigned. Google declined to renew. The contract moved to another vendor. https://www.pbs.org/newshour/show/amid-pressure-from-employees-google-drops-pentagons-project-maven-account

• Bulletin of the Atomic Scientists. “Was a Flying Killer Robot Used in Libya? Quite Possibly.” May 2021.

  The Kargu-2 in autonomous mode. Additional sourcing for the Libya drone incident. https://thebulletin.org/2021/05/was-a-flying-killer-robot-used-in-libya-quite-possibly/

• Lieber Institute at West Point. “Gospel, Lavender, and the Law of Armed Conflict.” 2024.

  West Point’s legal analysis of AI-directed targeting in Gaza. Additional sourcing for Lavender. https://lieber.westpoint.edu/gospel-lavender-law-armed-conflict/

• OHCHR. “Gaza: UN Experts Deplore Use of Purported AI to Commit ‘Domicide.’” April 2024.

  UN human rights experts on AI-directed destruction of civilian homes. https://www.ohchr.org/en/press-releases/2024/04/gaza-un-experts-deplore-use-purported-ai-commit-domicide-gaza-call

• Modern War Institute at West Point. “What Is ‘Meaningful Human Control’ Anyway?” 2024.

  No agreed definition. 20 seconds rubber-stamping an AI recommendation is technically “in the loop.” https://mwi.westpoint.edu/what-is-meaningful-human-control-anyway-cracking-the-code-on-autonomous-weapons-and-human-judgment/

• Defense One. “SecDef: China Is Exporting Killer Robots to the Mideast.” November 2019.

  China exports drones with lethal autonomy while calling for a treaty banning their use but not development. https://www.defenseone.com/technology/2019/11/secdef-china-exporting-killer-robots-mideast/161100/

• Newsweek. “Laura Nolan on Killer Robots and Project Maven.” 2019.

  “I was still part of the kill chain.” Google SRE who quit over Maven. Additional sourcing. https://www.newsweek.com/google-project-maven-artificial-intelligence-laura-nolan-killer-robots-department-defense-1459358

• Rolling Stone. “Did the U.S. Government Learn Nothing From WarGames?” 2026.

  Hegseth vs. Anthropic as WarGames made real. Additional sourcing for Pentagon dispute. https://www.rollingstone.com/tv-movies/tv-movie-features/war-games-anthropic-pete-hegseth-1235522766/

• Breaking Defense. “Insatiable Appetite for AI: Maven Usage Surged for Strikes on Iran, Pentagon AI Chief Says.” May 2026.

  Cameron Stanley, Pentagon CDAO: 13,000 targets in 38 days. Classified usage +89%. Daily token consumption +4,425%. Personnel showed an “insatiable appetite” for Maven. https://breakingdefense.com/2026/05/insatiable-appetite-for-ai-maven-usage-surged-for-strikes-on-iran-pentagon-ai-chief-says/

• Katrina Manson. “Iran War Tests Project Maven, US AI War Strategy.” Bloomberg, March 12, 2026.

  The closest US analog yet to the Abraham insider reporting on Lavender. Names Marine Colonel Drew Cukor as the figure who drove Maven from inside the Pentagon from 2017 forward. https://www.bloomberg.com/news/features/2026-03-12/iran-war-tests-project-maven-us-ai-war-strategy

• Time. “Iran School Strike Minab Tomahawk.” March 11, 2026.

  Shajareh Tayyebeh Elementary School, Minab, Hormozgan Province, destroyed on the first day of Operation Epic Fury. 156 dead, 120 schoolchildren. https://time.com/article/2026/03/11/iran-school-strike-minab-tomahawk/

• Amnesty International. “USA / Iran: Those Responsible for Deadly and Unlawful US Strike on School That Killed Over 100 Children Must Be Held Accountable.” March 2026.

  Western human-rights documentation of the Minab strike. Casualty figures sourced through Amnesty’s investigation, not Iranian government claims. https://www.amnesty.org/en/latest/news/2026/03/usa-iran-those-responsible-for-deadly-and-unlawful-us-strike-on-school-that-killed-over-100-children-must-be-held-accountable/

• OECD AI Incident Database. Entry 2026-04-14. April 14, 2026.

  Catalogues the Minab school strike as an MSS-attributable civilian-casualty incident. https://oecd.ai/en/incidents/2026-04-14-4ad0

• Rep. Jason Crow. “Crow, 120 Members Demand Answers on School Strike in Iran.” March 2026.

  Congressional response to the Minab strike. 120 House members + 46 senators led by Schatz, Shaheen, Van Hollen, Kaine, Warren on the parallel Senate letter. https://crow.house.gov/media/press-releases/crow-120-members-demand-answers-on-school-strike-in-iran

• Airwars. “The First Civilian Confirmed Killed in an AI-Assisted Strike.” December 2024.

  Abdul-Rahman al-Rawi, 20, Iraqi student killed February 2, 2024 in a CENTCOM strike wave. CENTCOM CTO Schuyler Moore confirmed Maven’s role in target identification. https://airwars.org/the-first-civilian-confirmed-killed-in-an-ai-assisted-strike/

Surveillance and Privacy

• TechCrunch. Signal’s Meredith Whittaker: AI Is Fundamentally a Surveillance Technology. September 2023.

  The refuser’s whole case in one headline; everything else is footnotes. https://techcrunch.com/2023/09/25/signals-meredith-whittaker-ai-is-fundamentally-a-surveillance-technology/

• TechCrunch. Signal President Meredith Whittaker Calls Out Agentic AI. March 2025.

  ‘Putting your brain in a jar,’ and handing the jar your bank login to save some typing. https://techcrunch.com/2025/03/07/signal-president-meredith-whittaker-calls-out-agentic-ai-as-having-profound-security-and-privacy-issues/

• Joy Buolamwini. Written Testimony, House Committee on Oversight & Reform. 22 May 2019.

  She measured the machine misreading Black women’s faces, then read the numbers into the record. https://docs.house.gov/meetings/GO/GO00/20190522/109521/HHRG-116-GO00-Wstate-BuolamwiniJ-20190522.pdf

• Jeremy Daum. Social Credit Resources. China Law Translate.

  The one primary that survives contact with the Black Mirror myth: it’s a court blacklist for deadbeats, not a national vibe score. https://www.chinalawtranslate.com/en/social-credit-resources/

• Court of Justice of the EU. SCHUFA Holding (Scoring), Case C-634/21. December 2023.

  Germany’s credit oracle scored you in the dark for decades; the CJEU ruled the black box counts as automated individual decision-making under GDPR Art. 22. https://curia.europa.eu/juris/document/document.jsf?docid=280426

• European Union. Corporate Sustainability Reporting Directive (Directive (EU) 2022/2464). Official Journal, 2022.

  ESG disclosure for firms, formalized — a behavioral report card for corporations, the same machinery pointed one rung up the ladder. https://eur-lex.europa.eu/eli/dir/2022/2464/oj

• FICO. FICO Score.

  The number that has quietly gated American credit for a generation, now folding in trended history. Population-scale scoring nobody voted for and nobody can audit. https://www.fico.com/en/products/fico-score

• autoevolution. How General Motors Can Remotely Disable a Stolen Car.

  ~1,500 slowdowns a month, for now requiring a police blessing — the ‘for now’ is the whole book. https://www.autoevolution.com/news/how-general-motors-can-remotely-disable-a-stolen-car-224715.html

• Jalopnik. Tesla Remotely Removes Autopilot Features From Customer Who Bought Car at Auction. February 2020.

  Car in the driveway, keys in hand, $8,000 of features deleted overnight. The owner owned the metal; Tesla owned the ‘yes.’ https://www.jalopnik.com/tesla-remotely-removes-autopilot-features-from-customer-1841472617/

• TechSpot. BMW Admits Heated Seat Subscriptions Were a Mistake.

  If they can rent you back the seat warmer already wired into your car, the engine is the same command with a different flag. https://www.techspot.com/news/111208-bmw-admits-heated-seat-subscriptions-mistake-but-commits.html

• MERICS. China’s Social Credit Score — Untangling Myth From Reality.

  The sober version: it’s purchase bans, not remote kill — a distinction held up entirely by policy, not architecture. https://merics.org/en/comment/chinas-social-credit-score-untangling-myth-reality

• European Commission. eCall: Emergency Assistance for Vehicles.

  An always-on cellular modem with GPS in every new car, that promises to only phone home in a crash. The promise is policy; the modem is architecture. https://europa.eu/youreurope/citizens/travel/security-and-emergencies/emergency-assistance-vehicles-ecall/index_en.htm

• Edmunds. GM Killed the Program That Sold Driving Data to Insurance Companies. March 2024.

  Terminated prospectively. The data already sold wasn’t recalled and the premiums already raised weren’t lowered. https://web.archive.org/web/20260706133934/https://www.edmunds.com/car-news/gm-killed-program-that-sold-driving-data-to-insurance-companies.html

• Consumer Federation of America. Auto Insurance Telematics White Paper. 2021.

  The primary on how ‘we only observe your driving’ becomes the pipeline that already runs to third parties. https://web.archive.org/web/20251211003422/https://consumerfed.org/wp-content/uploads/2021/05/Insurance-Auto-Telematics-White-Paper-5-26-21.pdf

• Comparitech. The World’s Most Surveilled Cities.

  The per-capita league table where China’s camera count laps the field. https://www.comparitech.com/vpn-privacy/the-worlds-most-surveilled-cities/

• Internet Freedom Foundation. Project Panoptic.

  A live tracker of India’s facial-recognition rollout — the deployments no statute ever authorized. https://panoptic.in/

• European Union. Regulation (EU) 2024/1689 (Artificial Intelligence Act). Official Journal, 2024.

  Bans real-time remote biometric ID in public, then writes the doors back in for serious crime, missing persons, and imminent threat. https://eur-lex.europa.eu/eli/reg/2024/1689/oj

• Joy Buolamwini and Timnit Gebru. Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification. PMLR (FAccT), 2018.

  The actual FR-bias primary: error up to 34.7% for darker-skinned women vs 0.8% for lighter-skinned men. Not ProPublica — that was COMPAS. https://proceedings.mlr.press/v81/buolamwini18a.html

• PCLOB. Executive Order 12333 Public Capstone Report. 2024.

  The oversight board’s own verdict on the authority most NSA collection actually runs under: it “has never been subject to meaningful oversight by either courts or Congress.” The watchdog admits it was never allowed in the yard. https://documents.pclob.gov/prod/Documents/OversightReport/b11b78e0-019f-44b9-ae4f-60e7eebe8173/12333%20Public%20Capstone.pdf

• Ronald Reagan. Executive Order 12333 — United States Intelligence Activities. December 4, 1981.

  The Reagan signature that still governs the majority of NSA surveillance — overseas, warrantless, and, by design, judicially unreviewable. https://www.archives.gov/federal-register/codification/executive-order/12333.html

• Stanford Law Review. Is the Foreign Intelligence Surveillance Court Really a Rubber Stamp? (online).

  The steelman of the 99.97% number — the government self-selects strong applications — which turns out to concede the point: a court that only ever sees pre-approved requests isn’t a check, it’s a co-author. https://www.stanfordlawreview.org/online/is-the-foreign-intelligence-surveillance-court-really-a-rubber-stamp/

• Center for Democracy & Technology. FISA Section 702 Issue Brief: FBI’s Misuse of FISA 702.

  The receipts behind the 278,000 non-compliant queries, and the prediction in the title: the procedural reforms will not break the pattern. https://cdt.org/insights/fisa-section-702-issue-brief-fbis-misuse-of-fisa-702/

• The White House. Memorandum on Advancing United States Leadership in Artificial Intelligence to Achieve National Security Objectives. October 24, 2024.

  The primary source for the NSA-to-AI handoff: the memo ordering a classified AI-model testing capability inside 120 days. The surveillance stack gets a machine-learning front end by executive order. https://bidenwhitehouse.archives.gov/briefing-room/presidential-actions/2024/10/24/memorandum-on-advancing-the-united-states-leadership-in-artificial-intelligence/

• DefenseScoop. DOD Raises Palantir Maven Smart System Contract to $1B+. May 23, 2025.

  The billion-dollar-plus number under the “Project Maven” claim — the CIA-seeded startup graduating to prime defense contractor. https://defensescoop.com/2025/05/23/dod-palantir-maven-smart-system-contract-increase/

• Darren Byler. Terror Capitalism: Uyghur Dispossession and Masculinity in a Chinese City. Duke University Press, 2022.

  An anthropologist watches his friends get surveilled, dispossessed, and disappeared over seven years of fieldwork. https://www.dukeupress.edu/terror-capitalism

  • Amazon: https://www.amazon.com/dp/1478017643

• NIST. “Face Recognition Vendor Test: Demographic Effects.” December 2019.

  189 algorithms, 18.27 million images. False positives 10-100x higher for Black faces. Eight wrongful arrests — all victims Black. https://www.nist.gov/news-events/news/2019/12/nist-study-evaluates-effects-race-age-sex-face-recognition-software

• The Markup. “Predictive Policing Software Is Terrible at Predicting Crimes.” October 2023.

  Less than 0.5% success rate. “Predicts future policing, not future crime.” https://themarkup.org/prediction-bias/2023/10/02/predictive-policing-software-terrible-at-predicting-crimes

• Edward Snowden disclosures. The Guardian, Washington Post, June 2013.

  2014 Pulitzer Prize. Tech companies encrypted data center links. Congress passed the USA FREEDOM Act. Everything else continued. https://snowden.glendon.yorku.ca/

• Washington Post. “NSA Infiltrates Links to Yahoo, Google Data Centers Worldwide.” October 30, 2013.

  The MUSCULAR program. The hand-drawn smiley face. Two Google engineers “exploded in profanity.” https://www.washingtonpost.com/world/national-security/nsa-infiltrates-links-to-yahoo-google-data-centers-worldwide-snowden-documents-say/2013/10/30/e51d661e-4166-11e3-8b74-d89d714ca4dd_story.html

• EFF. “Inside the Invasive, Secretive ‘Bossware’ Tracking Workers.” June 2020.

  Coined the term. Keystroke logging, stealth webcam photos, GPS tracking. 70% employer adoption by 2024. https://www.eff.org/deeplinks/2020/06/inside-invasive-secretive-bossware-tracking-workers

• Vice. “Internal Documents Show Amazon’s Dystopian System for Tracking Workers Every Minute of Their Shifts.” February 2020.

  Two hours of “time off task” and you’re fired. The algorithm does the firing. No human review. https://www.vice.com/en/article/internal-documents-show-amazons-dystopian-system-for-tracking-workers-every-minute-of-their-shifts/

• Mozilla Foundation. “Your Driving Data for the Price of a Gumball.” 2024.

  84% of car brands sell driver data. GM sent 600+ drives to LexisNexis. Hyundai sold 1.7 million vehicles’ data for 61 cents each. https://www.mozillafoundation.org/en/privacynotincluded/articles/your-driving-data-for-the-price-of-a-gumball-what-we-learned-from-two-us-senators-bonkers-new-letter-to-the-ftc/

• EFF. “Police Use of Face Recognition Continues to Wrack Real-World Harms.” January 2025.

  Eight documented wrongful arrests. All publicly identified victims are Black. https://www.eff.org/deeplinks/2025/01/police-use-face-recognition-continues-wrack-real-world-harms

• Reuters / Fortune. “Amazon Scraps Secret AI Recruiting Tool That Showed Bias Against Women.” October 2018.

  Penalized “women’s chess club captain.” Downgraded all-women’s colleges. Amazon concluded the bias was unfixable. https://fortune.com/2018/10/10/amazon-ai-recruitment-bias-women-sexist/

• Fortune. “ICE Allegedly Uses Palantir Tool Tracking Medicaid Data.” January 2026.

  FALCON, ImmigrationOS, ELITE — “Google Maps for deportable people.” ICE/CMS data-sharing covers 80 million Medicaid patients. https://fortune.com/2026/01/26/ice-allegedly-uses-palantir-tool-tracking-medicaid-data/

• The Intercept. “Peter Thiel’s Palantir Was Used to Bust Relatives of Migrant Children.” May 2019.

  ICE used unaccompanied minors to target their parents. The children were bait. https://theintercept.com/2019/05/02/peter-thiels-palantir-was-used-to-bust-hundreds-of-relatives-of-migrant-children-new-documents-show/

• HRW. “China’s Algorithms of Repression.” May 2019.

  Over 40,000 cameras. The IJOP app aggregates blood type, height, electricity usage, package deliveries. Huawei tested a “Uyghur alert.” https://www.hrw.org/report/2019/05/01/chinas-algorithms-repression/reverse-engineering-xinjiang-police-mass

• NPR. “How Facial Recognition Led to a False Arrest in Michigan.” June 2020.

  Robert Williams. Arrested in front of daughters aged 2 and 5. Held 30 hours. “I hope you don’t think all Black men look alike.” https://www.npr.org/2020/06/24/882683463/the-computer-got-it-wrong-how-facial-recognition-led-to-a-false-arrest-in-michig

• Democracy Now. “Porcha Woodruff: False Facial Recognition Arrest.” August 2023.

  8 months pregnant. Six officers at 8 a.m. Handcuffed for carjacking. Held 11 hours. Began having contractions in jail. https://www.democracynow.org/2023/8/9/porcha_woodruff_false_facial_recognition_arrest

• Undark. “Michigan Unemployment Fraud Algorithm.” June 2020.

  MiDAS falsely accused 40,000 people. 93% wrong. Tax refunds seized, wages garnished. https://undark.org/2020/06/01/michigan-unemployment-fraud-algorithm/

• Indiana Supreme Court. Perdue v. Gargano, 2012.

  FSSA’s generic denial notices violated Fourteenth Amendment Due Process. The named plaintiff Sheila Perdue, deaf and severely disabled, was terminated from benefits after the call-center system could not process her phone interview. https://caselaw.findlaw.com/court/in-supreme-court/1596533.html

• ACLU of Indiana. “Indiana Supreme Court Ruling Provides Neediest Hoosiers.” 2012.

  The aftermath of the failed $1.37 billion IBM contract to automate Indiana’s eligibility determinations. Mass denials to applicants who had cooperated. https://www.aclu-in.org/news/indiana-supreme-court-ruling-provides-neediest-hoosiers/

• Government Technology. “IBM Owes Indiana $78M for Failed Welfare Automation.” October 2019.

  Reciprocal commercial litigation. The state got $78 million. The applicants whose benefits were wrongly denied got nothing. https://www.ipm.org/2019-10-31/ibm-owes-indiana-78m-failed-welfare-automation-124981

• Center for Democracy & Technology. “Challenging the Use of Algorithm-driven Decision-making in Benefits Determinations Affecting People with Disabilities.” October 2020.

  Arkansas ARChoices, Idaho Department of Health and Welfare, Indiana FSSA — the algorithmic-benefits failure pattern across multiple states. Bradley Ledgerwood from 56 weekly care hours to 32. https://cdt.org/wp-content/uploads/2020/10/2020-10-21-Challenging-the-Use-of-Algorithm-driven-Decision-making-in-Benefits-Determinations-Affecting-People-with-Disabilities.pdf

• AP / PBS. “DOJ Examining AI Screening Tool Used by PA Child Welfare Agency.” 2023.

  Allegheny Family Screening Tool. Disproportionately flags Black families. Families not allowed to know their scores. https://www.pbs.org/newshour/nation/ap-report-doj-examining-ai-screening-tool-used-by-pa-child-welfare-agency

• ASPI. “How Mass Surveillance Works in Xinjiang.” 2020.

  Every 500 meters: ID card, facial recognition, phone scan. Additional sourcing for Uyghur surveillance. https://xjdp.aspi.org.au/explainers/how-mass-surveillance-works-in-xinjiang/

• Byline Times. “Palantir’s Tools for Tracking Dissent.” January 2026.

  Palantir’s ELITE as “Google Maps for deportable people.” Additional sourcing for Palantir/ICE. https://bylinetimes.com/2026/01/29/domestic-terrorism-ice-contractor-palantirs-tools-for-tracking-dissent/

• Oxford Blavatnik School. “Australia’s Robodebt Scheme.” 2023.

  433,000 unlawful debts. 663 deaths. $1.2 billion repaid. Additional sourcing. https://www.bsg.ox.ac.uk/blog/australias-robodebt-scheme-tragic-case-public-policy-failure

• American Prospect. “Amazon Uses Arsenal of AI Weapons Against Workers.” March 2025.

  France fined Amazon 32 million euros for “excessively intrusive” monitoring. Additional sourcing. https://prospect.org/labor/2025-03-13-amazon-uses-arsenal-of-ai-weapons-against-workers/

• NCCPR Blog. “The Machine Has Labeled You as High Risk.” December 2023.

  Andrew and Lauren Hackney: baby taken by algorithm for two years. Additional sourcing for Allegheny. https://www.nccprblog.org/2023/12/the-machine-has-labeled-you-as-high-risk.html

• WPXI. “Federal Judge Clears Family’s Federal Lawsuit Against Allegheny County CYF to Move Forward.” April 2026.

  Hackney v. Allegheny County moves past motion-to-dismiss. The algorithm took their baby for two years; now the lawsuit advances. https://www.wpxi.com/news/local/federal-judge-clears-familys-federal-lawsuit-against-allegheny-county-cyf-move-forward/7UTKL5DNSBGALBEW3MT57QGUFI/

• ACLU. “Family Surveillance by Algorithm: The Rapidly Spreading Tools Few Have Heard Of.” 2021 (updated through 2026).

  26 states deploying comparable child-welfare risk-scoring systems by mid-2026. The Hackney pattern, national. https://www.aclu.org/wp-content/uploads/document/2021.09.28_Family_Surveillance_by_Algorithm.pdf

• NPR. “Oregon Drops Artificial Intelligence Child Abuse Cases.” June 2022.

  Oregon DHS abandoned its Allegheny-analog tool after audit. The pattern of abandonment alongside the pattern of expansion. https://www.npr.org/2022/06/02/1102661376/oregon-drops-artificial-intelligence-child-abuse-cases

• EachOther. “Councils Using Data and Algorithms in Child Protection.” November 2019.

  London Borough of Hackney and other UK councils using comparable predictive tools. Additional sourcing for international parallel. https://eachother.org.uk/councils-using-data-and-algorithms-in-child-protection/

• Reason. “Child Welfare Algorithm May Unfairly Target Disabled Parents.” February 2023.

  DOJ ADA complaints against the Allegheny tool. Additional sourcing. https://reason.com/2023/02/02/child-welfare-algorithm-may-unfairly-target-disabled-parents-complaints-to-doj-allege/

• Michigan Auditor General. “Unemployment Insurance Agency Fraud Investigation.” 2016 (PDF).

  93% of MiDAS fraud determinations were wrong. Additional sourcing. https://audgen.michigan.gov/finalpdfs/15_16/r641059315.pdf

• Bridge Michigan. “Broken: The Human Toll of Michigan’s Unemployment Fraud Saga.” 2020.

  Carmelita Colvin denied a job over an erroneous algorithmic charge. Additional sourcing for MiDAS. https://bridgemi.com/michigan-government/broken-human-toll-michigans-unemployment-fraud-saga/

• Regulatory Oversight. “Clearview AI $51.75M Settlement.” April 2025.

  ~30 billion scraped faceprints (CEO Hoan Ton-That, 2023; 100-billion goal stated). $51.75M settlement. Additional sourcing for Clearview. https://www.regulatoryoversight.com/2025/04/51-75m-settlement-in-clearview-ai-biometric-privacy-litigation-illustrates-creative-resolution-for-startups-facing-parallel-litigation-and-enforcement-action/

• Scroll.in. “Denied Food Because She Did Not Have Aadhaar-Linked Ration Card, Jharkhand Girl Dies of Starvation.” 2017.

  Santoshi Kumari, age 11. Eight days without food. New ration card arrived two weeks after her death. https://scroll.in/article/854225/denied-food-because-she-did-not-have-aadhaar-linked-ration-card-jharkhand-girl-dies-of-starvation

• OHCHR. “Assessment of Human Rights Concerns in Xinjiang.” August 2022.

  “May constitute international crimes, in particular crimes against humanity.” Published on Bachelet’s last day in office. https://www.ohchr.org/en/documents/country-reports/ohchr-assessment-human-rights-concerns-xinjiang-uyghur-autonomous-region

• ICIJ. “China Cables.” November 2019.

  Operational manual for the camps. IJOP flagged ~24,000 people in one week. ~15,000 sent to camps. https://www.icij.org/investigations/china-cables/

• Xinjiang Police Files. May 2022.

  Direct photos of detainees — children aged 15, the oldest 73. Shoot-to-kill orders. Direct photos of detainees — the youngest 15, the oldest 73. Shoot-to-kill orders. https://www.xinjiangpolicefiles.org/

• Carnegie Endowment. “The Global Expansion of AI Surveillance.” September 2019.

  Chinese AI surveillance tech supplied to 63+ countries. Huawei alone: 50+ countries. https://carnegieendowment.org/research/2019/09/the-global-expansion-of-ai-surveillance

• PCLOB. “Report on Section 702.” 2023.

  278,000+ non-compliant FBI queries. Searched BLM protesters, Jan 6 suspects, journalists, a sitting member of Congress. https://documents.pclob.gov/prod/Documents/OversightReport/054417e4-9d20-427a-9850-862a6f29ac42/2023%20PCLOB%20702%20Report%20(002).pdf

• TechCrunch. “FBI Recommends Encrypted Messaging Apps to Combat Chinese Hackers.” December 2024.

  The FBI that spent years trying to break encryption now says use Signal. Salt Typhoon exploited CALEA backdoors. https://techcrunch.com/2024/12/04/fbi-recommends-encrypted-messaging-apps-combat-chinese-hackers/

• European Union. Regulation (EU) 2024/1183 (eIDAS 2.0), European Digital Identity Framework. In force May 20, 2024.

  The law that turns a wallet on your phone into the front door of European life: every member state must offer one by end-2026, every relying party must accept it by late 2027. https://eur-lex.europa.eu/eli/reg/2024/1183/oj/eng

• Supreme Court of the United States. Free Speech Coalition, Inc. v. Paxton. 606 U.S. ___ (June 27, 2025).

  6-3, intermediate scrutiny: age-verification for “harmful to minors” content survives the First Amendment. Kagan, in dissent, wanted strict scrutiny. The wedge is now constitutional. https://supreme.justia.com/cases/federal/us/606/23-1122/

• ACLU. Kashem v. Barr (no-fly-list due-process challenge).

  A federal court called the process unconstitutional; the government’s fix still will not tell you why you are on it. https://www.aclu.org/cases/kashem-et-al-v-barr-et-al-aclu-challenge-government-no-fly-list

• Congressional Research Service. Terrorist Databases and the No Fly List (R43730).

  The watchlist architecture, described by Congress’s own analysts: secret criteria, no exit, and no way in to contest. https://www.everycrsreport.com/reports/R43730.html

• World Economic Forum. The Known Traveller Digital Identity Concept. 2018.

  Blockchain biometric travel ID, pitched at Davos, stalled by COVID and never revived — the trial balloon that popped. https://www3.weforum.org/docs/WEF_The_Known_Traveller_Digital_Identity_Concept.pdf

• World Bank. Identification for Development (ID4D).

  850 million people without ID, 60 countries, funded by Gates / UK / France / NORAD / Omidyar. Universal enrollment as development aid. https://id4d.worldbank.org/

• House of Commons Library. Digital ID in the UK (CBP-10369).

  Parliament’s own librarians on the scheme three million people petitioned against. https://commonslibrary.parliament.uk/research-briefings/cbp-10369/

• Electronic Frontier Foundation. Obama Expands Surveillance Powers on His Way Out. EFF, January 2017.

  The reformer’s parting gift: a farewell rule loosening who inside the government gets to see the raw take, delivered just in time for the next administration to inherit it. https://www.eff.org/deeplinks/2017/01/obama-expands-surveillance-powers-his-way-out

• Electronic Frontier Foundation. U.S. Senate and Biden Administration Shamefully Renew and Expand FISA Section 702. EFF, April 2024.

  The candidate who once called 702 a civil-liberties violation signs its two-year expansion: EFF’s tally of the 60-34 midnight capitulation and everything it broadened. https://www.eff.org/deeplinks/2024/04/us-senate-and-biden-administration-shamefully-renew-and-expand-fisa-section-702-0

• The Intercept. Obama Opens NSA’s Vast Trove of Warrantless Data to Entire Intelligence Community, Just in Time for Trump. The Intercept, 13 January 2017.

  The primary reporting on the January 2017 rule: the NSA’s raw, unminimized haul opened to all sixteen agencies, the masking of American identities dropped on the way out the door. https://theintercept.com/2017/01/13/obama-opens-nsas-vast-trove-of-warrantless-data-to-entire-intelligence-community-just-in-time-for-trump/

• Pam Dixon. A Failure to “Do No Harm”: India’s Aadhaar Biometric ID Program and Its Inability to Protect Privacy in Relation to Measures in Europe and the U.S. Health and Technology, 2017.

  The peer-reviewed autopsy of Aadhaar’s privacy design, written before the body count came in: biometric enrollment as a right without a corresponding right to say no. https://pmc.ncbi.nlm.nih.gov/articles/PMC5741784/

• Economic and Political Weekly. Aadhaar Failures: A Tragedy of Errors. EPW Engage, 2018.

  The failure rates in one place: 49% authentication failure in Jharkhand, 37% in Rajasthan, the exclusion the architects call leakage prevention. https://www.epw.in/engage/article/aadhaar-failures-food-services-welfare

• Institute for Government. Digital ID Cards. Institute for Government explainer, 2025.

  Whitehall’s own think tank laying out the BritCard, its Estonian model, and the infrastructure that advanced while the mandate was quietly dropped. https://www.instituteforgovernment.org.uk/explainer/digital-id-cards

• Congressional Research Service. Supreme Court Upholds State Age-Verification Requirement for Certain Websites (LSB11354). CRS, 2025.

  Congress’s lawyers on Free Speech Coalition v. Paxton: intermediate scrutiny wins, and the age-check wedge is now precedent the states can build on. https://www.everycrsreport.com/reports/LSB11354.html

Health Data and Surveillance

• U.S. Federal Bureau of Investigation. Chinese Hackers Charged in Equifax Breach. FBI, February 2020.

  Four PLA members of the 54th Research Institute, 145 million Americans, one unpatched Apache Struts bug — indicted knowing Beijing will never extradite. https://www.fbi.gov/news/stories/chinese-hackers-charged-in-equifax-breach-021020

• CISA/NSA/FBI. PRC State-Sponsored Actors Compromise and Maintain Persistent Access to U.S. Critical Infrastructure (AA24-038A). 2024.

  Volt Typhoon sat inside the infrastructure ‘for at least five years,’ pre-positioning for the day the lights need to go out. Living off the land, so there’s nothing to scan for. https://www.cisa.gov/news-events/cybersecurity-advisories/aa24-038a

• HIPAA Journal. Change Healthcare Responding to Cyberattack.

  One legacy Citrix box, no MFA, one-third of US claims traffic — 192.7 million people and $2.87 billion later, the bill for a missing second factor. https://www.hipaajournal.com/change-healthcare-responding-to-cyberattack/

• World Health Assembly. World Health Assembly Adopts Historic Pandemic Agreement. WHO, 20 May 2025.

  Three years of negotiation for a treaty that cannot enter into force until an annex nobody has finished gets finished. The ratchet’s paperwork phase. https://www.who.int/news/item/20-05-2025-world-health-assembly-adopts-historic-pandemic-agreement-to-make-the-world-more-equitable-and-safer-from-future-pandemics

• World Health Organization. Amended International Health Regulations Enter Into Force. WHO, 19 September 2025.

  The binding half actually locked in — eleven states opted out, the other 185 bound by a ‘pandemic emergency’ trigger they can’t vote down after the fact. https://www.who.int/news/item/19-09-2025-amended-international-health-regulations-enter-into-force

• Gavi, the Vaccine Alliance. The Gates Foundation — Donor Profile.

  Gavi’s own books: $7.7 billion from a single foundation, its largest grantee, no lobbyist ever registered because it’s a grant, not a bribe. https://www.gavi.org/investing-gavi/funding/donor-profiles/gates-foundation

• Coalition for Epidemic Preparedness Innovations. CEPI Officially Launched. 2017.

  Davos, 2017: $460M from Gates, Wellcome, and three governments to build the vaccine pipeline they’d own the priorities of. Announced from a ski resort. https://cepi.net/cepi-officially-launched

• World Health Organization. How WHO Is Funded.

  84% of the budget is voluntary and mostly earmarked, which is a polite way of saying the donors write the program, not the member states. https://www.who.int/about/funding

• Oliver J. Watson et al. Global Impact of the First Year of COVID-19 Vaccination. The Lancet Infectious Diseases, 2022.

  14.4 to 19.8 million deaths averted in twelve months — the single largest life-saving line item the captured institutions can point to, and it’s real. https://www.imperial.ac.uk/news/237591/vaccinations-have-prevented-almost-20-million/

• U.S. Department of State. PEPFAR Latest Global Results Factsheet (Dec. 2024). 2024.

  26 million lives, $120 billion, the largest single-disease commitment any nation has made — run by the same federal bureaucracy the rest of the chapter indicts. https://www.state.gov/pepfar-latest-global-results-factsheet-dec-2024/

• Ivan Illich. Medical Nemesis: The Expropriation of Health. Calder & Boyars, 1975.

  The book that named iatrogenesis and ‘radical monopoly’ fifty years before ‘follow the science’ made his thesis a headline. https://en.wikipedia.org/wiki/Medical_Nemesis

• Erin Banco et al. How Bill Gates and Partners Used Their Clout to Control the Global Covid Response. POLITICO, 2022.

  Seven-month investigation into four Gates-linked NGOs steering the COVID response with, per the subtitle, ’little oversight.’ https://siidata.org/how-bill-gates-and-partners-used-their-clout-to-control-the-global-covid-response-with-little-oversight-politico/

• CNN (Nectar Gan et al.). China’s Bank Run Victims Planned to Protest. Then Their Covid Health Codes Turned Red. June 2022.

  The health pass proved its second use the moment a green code went red at the city limits to keep depositors away from their own frozen money. https://www.cnn.com/2022/06/15/china/china-zhengzhou-bank-fraud-health-code-protest-intl-hnk

• National Health Authority (India). Ayushman Bharat Digital Mission.

  A health ID for a billion-plus people, sold as continuity-of-care and built on the Aadhaar rail — the exact asset that got repurposed elsewhere. https://abdm.gov.in/

• U.S. Department of Health and Human Services. Disclosures for Law Enforcement Purposes (HIPAA).

  HIPAA’s own text is where the law-enforcement door is written — disclosure without consent, under a menu of conditions. https://www.hhs.gov/hipaa/for-professionals/special-topics/disclosures-law-enforcement/index.html

• Human Rights Watch. China: Minority Region Collects DNA from Millions. December 13, 2017.

  18.8 million Xinjiang residents fingerprinted, iris-scanned, and blood-typed under a free-health-check banner. Thermo Fisher supplied the sequencers. https://www.hrw.org/news/2017/12/13/china-minority-region-collects-dna-millions

• Human Rights Watch. China: Voice Biometric Collection Threatens Privacy. October 22, 2017.

  Voiceprints harvested with iFlytek, which makes 80% of the country’s speech tech. The voice line does not belong to the DNA report. https://www.hrw.org/news/2017/10/22/china-voice-biometric-collection-threatens-privacy

• International Civil Aviation Organization. Doc 9303 — Machine Readable Travel Documents (8th ed.). ICAO, 2021.

  The UN standard that put a face, a fingerprint, and an iris on a chip in every e-passport. https://www.icao.int/publications/doc-series/doc-9303

• Unique Identification Authority of India. Biometrics Committee Report. UIDAI, 2010.

  The spec that decided a billion-plus bodies would be keyed on ten prints and two irises. https://uidai.gov.in/en/media-resources/uidai-documents/circulars-memorandums-notification/reports/7091-biometrics-committee.html

• U.S. Department of Homeland Security / CBP. PIA-056: Traveler Verification Service. DHS, 2018 (updated 2021).

  The paperwork that turned ‘we already have your passport photo’ into a camera at the exit gate for 200M+ travelers. https://www.dhs.gov/publication/dhscbppia-056-traveler-verification-service

• Electronic Frontier Foundation. Cell Phone Searches.

  The device-seizure-at-the-border leg of biometric collection. https://www.eff.org/issues/cell-phone-searches

• FTC. FTC Finalizes Order with Flo Health. June 2021.

  The world’s most popular period tracker treated ‘we won’t share your data’ as ad copy. The order carried no fine. https://www.ftc.gov/news-events/news/press-releases/2021/06/ftc-finalizes-order-flo-health-fertility-tracking-app-shared-sensitive-health-data-facebook-google

• FTC. FTC to Ban BetterHelp from Revealing Consumers’ Data. March 2023.

  Therapy-intake answers piped to Facebook, Snapchat, Criteo, and Pinterest. The $7.8M penalty was a rounding error. https://www.ftc.gov/news-events/news/press-releases/2023/03/ftc-ban-betterhelp-revealing-consumers-data-including-sensitive-mental-health-information-facebook

• Joanne Kim (Duke Sanford). Data Brokers and the Sale of Americans’ Mental Health Data. February 2023.

  Eleven of twenty-six brokers would sell depression, ADHD, and anxiety records. Five and a half cents a breakdown. https://techpolicy.sanford.duke.edu/data-brokers-and-the-sale-of-americans-mental-health-data/

• ProPublica. Federal Patient Privacy Law Does Not Cover Most Period-Tracking Apps. 2022.

  HIPAA stops at the clinic door; the app in your pocket is on the other side of it. https://www.propublica.org/article/period-app-privacy-hipaa

• NPR. 23andMe Files for Bankruptcy: What It Means for Your Genetic Data. March 2025.

  Fifteen million spit-tubes become a line item in a Chapter 11 filing. The privacy policy always said they could. https://www.npr.org/2025/03/24/nx-s1-5338622/23andme-bankruptcy-genetic-data-privacy

• Office of the California Attorney General. Consumer Alert for 23andMe Customers. March 2025.

  A state AG telling citizens to delete their DNA before the auction. https://oag.ca.gov/news/press-releases/attorney-general-bonta-urgently-issues-consumer-alert-23andme-customers

• Joseph Cox (Vice). Data Broker Is Selling Location Data of People Who Visit Abortion Clinics. 2022.

  A week of visits to 600+ Planned Parenthood sites, $160. Anonymization by rounding error. https://www.vice.com/en/article/location-data-abortion-clinics-safegraph-planned-parenthood/

• Federal Trade Commission. FTC v. Kochava Inc.

  The government’s attempt to stop a broker from selling the map of who went to a clinic, a shelter, or a church. https://www.ftc.gov/legal-library/browse/cases-proceedings/ftc-v-kochava-inc

• Mozilla Foundation. Top Mental Health and Prayer Apps Fail Spectacularly at Privacy. 2022.

  29 of 32 flunked. Mozilla called it the worst category it had ever reviewed. https://www.mozillafoundation.org/en/blog/top-mental-health-and-prayer-apps-fail-spectacularly-at-privacy-security/

• NBC News. Priest Outed via Grindr App Highlights Rampant Data Tracking. 2021.

  If commercial location data can burn the general secretary of the US bishops, it can burn anyone. No warrant, no badge. https://www.nbcnews.com/tech/security/priest-outed-grindr-app-highlights-rampant-data-tracking-rcna1493

• Datatilsynet (Norwegian DPA). Record Fine to Grindr Confirmed. 2023.

  EUR 6.5M for handing GPS, profiles, and sexual orientation to five ad firms. The appeal failed. https://www.datatilsynet.no/en/news/aktuelle-nyheter-2023/record-fine-grindr-confirmed/

• EFF. Maryland and Montana Pass First Laws Restricting Law Enforcement Access to Genetic Genealogy Databases. June 2021.

  Two states drew a line around forensic DNA searches. Forty-eight did not. https://www.eff.org/deeplinks/2021/06/maryland-and-montana-pass-nations-first-laws-restricting-law-enforcement-access

• NPR. In Hunt for Golden State Killer, Investigators Uploaded His DNA to a Genealogy Site. April 2018.

  The case that proved your relatives’ spit is your exposure. You never tested; your cousin did. https://www.npr.org/sections/thetwo-way/2018/04/27/606624218/in-hunt-for-golden-state-killer-investigators-uploaded-his-dna-to-genealogy-site

• EPIC. Two Years Post-Dobbs: A Commercial Surveillance Landscape Harmful to Abortion Seekers. 2024.

  How shield laws die: a prosecutor doesn’t subpoena the period app when a broker will just sell the same data. https://epic.org/two-years-post-dobbs-a-commercial-surveillance-landscape-that-is-confusing-complicated-and-harmful-to-abortion-seekers/

• HIPAA Journal. Healthcare Data Breach Statistics.

  The running scoreboard: 2024 breached ~277 million records, north of 80% of the country. https://www.hipaajournal.com/healthcare-data-breach-statistics/

• NIH Director Francis Collins email (FOIA). October 2020.

  To Fauci, re Great Barrington Declaration: “there needs to be a quick and devastating published take down.” https://www.documentcloud.org/documents/21066874-nih-foia-emails

• The Lancet. “Statement in Support of Scientists Combatting COVID-19.” February 2020.

  Organized by Peter Daszak, whose EcoHealth Alliance funded gain-of-function research at Wuhan. Undisclosed conflict of interest. https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(20)30418-9/fulltext

• OpenSecrets. “Pharmaceutical / Health Products Lobbying.” 2023.

  $374M in 2023. #1 industry. 1,800 lobbyists — 3.4 per member of Congress. https://www.opensecrets.org/federal-lobbying/industries

• WHO. “Launch of the Global Digital Health Certification Network.” June 2023.

  EU Digital COVID Certificate became permanent WHO global system. 51 countries. “The first building block.” https://www.who.int/news/item/05-06-2023-the-european-commission-and-who-launch-landmark-digital-health-initiative-to-strengthen-global-health-security

• ORCHA (Organisation for the Review of Care and Health Apps). “Period Tracker Report.” 2022.

  84% of 25 period tracking apps shared data with third parties — marketing (68%), research (40%), developer services (40%). Only 1 of 25 used explicit in-app consent. Your most intimate biological data, monetized before you finish your first cycle. https://us.orchahealth.com/84-of-25-womens-health-period-tracker-apps-share-data-with-third-parties/ PDF: https://25077478.fs1.hubspotusercontent-eu1.net/hubfs/25077478/ORCHA%20-%20Period%20Tracker%20Report%20UK_5Jan2023.pdf

• CNBC. “Startup Orchid Launches Test to Identify Genetic Defects in IVF Embryos.” December 2023.

  Noor Siddiqui’s polygenic embryo screening service. $2,500 per embryo. First baby December 2023. Schizophrenia, Alzheimer’s, type 2 diabetes — ranked and selected. https://www.cnbc.com/2023/12/05/startup-orchid-launches-test-to-identify-genetic-defects-in-ivf-embryos.html

• MIT Technology Review. “The World’s First Gattaca Baby Tests Are Finally Here.” November 2019.

  Genomic Prediction / LifeView. Stephen Hsu’s startup. Hsu later resigned from Michigan State (June 2020) after protests over his work. https://www.technologyreview.com/2019/11/08/132018/polygenic-score-ivf-embryo-dna-tests-genomic-prediction-gattaca/

• Lior Pachter. “The Amoral Nonsense of Orchid’s Embryo Selection.” liorpachter.wordpress.com, April 2021.

  Caltech computational biologist on the predictive-validity gap in polygenic embryo scores. Technical pushback from someone who can do the math. https://liorpachter.wordpress.com/2021/04/12/the-amoral-nonsense-of-orchids-embryo-selection/

• Hannah Devlin / HOPE not hate. “British Company Offering Wealthy Couples Embryo Screening for IQ.” The Guardian, October 18, 2024.

  Heliospect / PolygenX caught in HOPE not hate sting. $50,000 service offering “approximately six IQ points” using UK Biobank data accessed June 2023. https://www.theguardian.com/science/2024/oct/18/us-startup-charging-couples-to-screen-embryos-for-iq

• Washington Post. “Inside Orchid’s Polygenic Screening of Embryos.” July 2025.

  Update on Orchid’s expansion, customer base, founder Noor Siddiqui’s continuing pitch. https://www.washingtonpost.com/technology/2025/07/16/orchid-polygenic-screening-embryos-fertility/

• Yang et al. “AI Embryo Selection Versus Embryologist Scoring.” Nature Medicine, June 2024.

  Randomized clinical trial of Vitrolife’s iDAScore. No statistically significant improvement in pregnancy rates. Product still deployed at hundreds of IVF clinics.

• AP / PBS. “DOJ Examining AI Screening Tool Used by PA Child Welfare Agency.” 2023.

  DOJ Civil Rights Division Title II ADA probe opened November 2022. As of mid-2026, no public findings. https://www.pbs.org/newshour/nation/ap-report-doj-examining-ai-screening-tool-used-by-pa-child-welfare-agency

• Jean Drèze, Reetika Khera, Anmol Somanchi. “Balancing Corruption and Exclusion: Incidence of Welfare Programs in Jharkhand.” Economic & Political Weekly, 2020 (and subsequent published analyses).

  Aadhaar in welfare programs. Approximately 29% authentication-failure-driven disruption rate in PMMVY (maternity benefit) enrollment. The deduplication system worked exactly as designed; the women it was designed to serve did not.

• Office of the New York State Attorney General. Attorney General James Sues 23andMe to Protect New Yorkers’ Genetic Data. 2025.

  James and a bipartisan coalition of 27 other attorneys general suing to keep a bankrupt company’s genetic database from being sold like office furniture. https://ag.ny.gov/press-release/2025/attorney-general-james-sues-23andme-protect-new-yorkers-genetic-data

• Ryan Holthouse, Trevor Owens, Suman Bhunia. “The 23andMe Data Breach: Analyzing Credential Stuffing Attacks, Security Vulnerabilities, and Mitigation Strategies.” arXiv:2502.04303, 2025.

  A five-month credential-stuffing run nobody noticed, turning one reused password into 6.9 million people’s ancestry. The post-mortem on how you lose a genome database quietly. https://arxiv.org/abs/2502.04303

• J.B. Branch. “Privacy, Consent, and National Security After the 23andMe Bankruptcy.” Lawfare, July 17, 2025.

  The database changes hands and the CFIUS trigger does not fire, because the buyer was American. The receipt for “a genome is a strategic asset only when the wrong country wants it.” https://www.lawfaremedia.org/article/privacy–consent–and-national-security-after-the-23andme-bankruptcy

• Adam Klein. Prepared Testimony: The Privacy and National Security Implications of the 23andMe Bankruptcy. Senate Judiciary Committee, June 11, 2025.

  The genome as adversary intelligence, argued to Congress by the man who used to run the federal privacy oversight board. Population-level legibility, not the individual profile, is the payload. https://www.judiciary.senate.gov/imo/media/doc/61343549-a23b-e555-8c28-ce114cbfe8c8/2025-06-11%20-%20Testimony%20-%20Klein.pdf

• Todd Lencz et al. “Utility of Polygenic Embryo Screening for Disease Depends on the Selection Strategy.” eLife 10:e64716, 2021.

  The field publishing its own ceiling: lowest-risk-of-five buys ~50% relative risk reduction and under half a percentage point absolute. Same selection, two numbers; the marketing prints one. https://doi.org/10.7554/eLife.64716

• Patrick Turley et al. “Problems with Using Polygenic Scores to Select Embryos.” New England Journal of Medicine 385(1):78-86, 2021.

  The strongest paper against the practice, written from inside the field: predictive power collapses on embryos, and selecting against one trait can drag a correlated one along unnamed. https://www.nejm.org/doi/full/10.1056/NEJMsr2105065

• Grebe et al. “Clinical Utility of Polygenic Risk Scores for Embryo Selection: A Points to Consider Statement of the ACMG.” Genetics in Medicine, 2024.

  The College of Medical Genetics, on the record: the evidence is insufficient and the test “should not be offered as a clinical service.” Sold anyway. https://www.gimjournal.org/article/S1098-3600(23)01046-0/fulltext

• Li et al. “ImputePGTA: Accurate Embryo Genotyping and Polygenic Scoring from Ultra-Low-Pass Sequencing.” medRxiv, 2025.

  Herasight’s method paper — the genotyping engine under the “world’s most powerful cognitive predictor” marketing it will not let peer review grade first. https://doi.org/10.1101/2025.11.07.25339763

• Privacy International. The Looming Disaster of Immunity Passports and Digital Identity. Privacy International long-read, 2020.

  Written before the passports shipped: once biometric data are collected and filed, there is no delete button, the line the whole dossier turns on. https://privacyinternational.org/long-read/4074/looming-disaster-immunity-passports-and-digital-identity

• Mozilla Foundation. In Post-Roe v. Wade Era, Mozilla Labels 18 of 25 Popular Period and Pregnancy Tracking Apps with “Privacy Not Included” Warning. 2022.

  The reproductive-health category graded, and it failed: 18 of 25 flunked, with only Euki, which keeps everything on the phone, earning a pass. https://www.mozillafoundation.org/en/privacynotincluded/articles/in-post-roe-v-wade-era-mozilla-labels-18-of-25-popular-period-and-pregnancy-tracking-tech-with-privacy-not-included-warning/

• Federal Trade Commission. Proposed FTC Order Will Prohibit Telehealth Firm Cerebral from Using or Disclosing Sensitive Data for Advertising Purposes. April 2024.

  Nearly 3.2 million people’s therapy intake handed to LinkedIn, Snapchat, and TikTok via tracking pixels; the $10M penalty suspended after $2M for inability to pay. https://www.ftc.gov/news-events/news/press-releases/2024/04/proposed-ftc-order-will-prohibit-telehealth-firm-cerebral-using-or-disclosing-sensitive-data

• Federal Trade Commission. FTC Takes Action Against Gravy Analytics, Venntel for Unlawfully Selling Location Data Tracking Consumers. December 2024.

  The broker that mapped visits to clinics, churches, and shelters, told to stop: the 2024 sequel to the Kochava suit, same product, different defendant. https://www.ftc.gov/news-events/news/press-releases/2024/12/ftc-takes-action-against-gravy-analytics-venntel-unlawfully-selling-location-data-tracking-consumers

• New England Journal of Medicine. Bankruptcy, Genetic Information, and Privacy. NEJM Perspective, 2025.

  The medical establishment’s own journal conceding that existing law barely protects fifteen million genomes once the company holding them files Chapter 11. https://www.nejm.org/doi/full/10.1056/NEJMp2415835

• Pregnancy Justice. Pregnancy as a Crime: A Preliminary Report on the First Year After Dobbs. 2024.

  At least 210 pregnant people charged in twelve months, the highest single-year count ever recorded, 191 of them requiring no proof of harm to anyone. https://www.pregnancyjusticeus.org/resources/pregnancy-as-a-crime-a-preliminary-report-on-the-first-year-after-dobbs/

• ProPublica. Big Data + Big Pharma = Big Money. ProPublica, 2015.

  Three-quarters of US retail pharmacies feeding prescribing records into a $9 billion intelligence trade, so the drug rep already knows what the doctor writes before walking in. https://www.propublica.org/article/big-data-big-pharma-big-money

• Yaniv Erlich et al. Identity Inference of Genomic Data Using Long-Range Familial Searches. Science, 2018.

  The paper that did the arithmetic behind the Golden State Killer bust: about 60% of white Americans are already reachable through a cousin who spat in a tube. https://www.science.org/doi/10.1126/science.aau4832

• U.S. Department of Justice. Member of a China-Based Hacking Group Indicted for the 2015 Data Breach of Health Insurer Anthem Affecting Over 78 Million People. DOJ Office of Public Affairs, May 2019.

  Fujie Wang and a John Doe, indicted in absentia, for 78.8 million records lifted with Sakula: a receipt Beijing will never honor. https://www.justice.gov/archives/opa/pr/member-sophisticated-china-based-hacking-group-indicted-series-computer-intrusions-including

• U.S. Department of Health and Human Services, Office for Civil Rights. Anthem Pays OCR $16 Million in Record HIPAA Settlement Following Largest U.S. Health Data Breach in History. HHS, 2018.

  The largest HIPAA settlement ever, and still a rounding error against 78.8 million exfiltrated records: the price of losing a country’s medical files. https://www.hhs.gov/hipaa/for-professionals/compliance-enforcement/agreements/anthem/index.html

• U.S. Department of Justice. Chinese Military Personnel Charged with Computer Fraud, Economic Espionage and Wire Fraud for Hacking into Equifax. DOJ Office of Public Affairs, February 2020.

  Four PLA 54th Research Institute officers named for the Equifax haul: the indictment that says the quiet part, state military, not lone criminals. https://www.justice.gov/archives/opa/pr/chinese-military-personnel-charged-computer-fraud-economic-espionage-and-wire-fraud-hacking

• U.S. Department of Justice. Seven International Cyber Defendants, Including APT41 Actors, Charged in Connection with Computer Intrusion Campaigns Against More Than 100 Victims Globally. DOJ Office of Public Affairs, September 2020.

  Five Chinese nationals for 100-plus institutions across the healthcare map, MSS-linked and untouchable: Winnti with a badge. https://www.justice.gov/archives/opa/video/seven-international-cyber-defendants-including-apt41-actors-charged-connection-computer

• U.S. HHS Health Sector Cybersecurity Coordination Center (HC3). China-Based Threat Actor Profiles. HHS HC3, August 2023.

  The government’s own threat brief naming APT41 and APT10 as deliberate hunters of US medical IP and public-health data. https://www.hhs.gov/sites/default/files/china-based-threat-actor-profiles-tlpclear.pdf

• U.S. HHS Health Sector Cybersecurity Coordination Center (HC3). Data Exfiltration Trends in Healthcare. HHS HC3.

  Espionage is nearly one in six healthcare breaches: the sector graded on how much of it walks out the door and where it goes. https://www.hhs.gov/sites/default/files/data-exfiltration-in-healthcare-tlpclear.pdf

• Recorded Future (Insikt Group). Dark Covenant 3.0: Controlled Impunity and Russia’s Cybercriminals. Recorded Future, 2023.

  The evolution from passive tolerance to active management: never attack Russia, do favors on demand, the state-criminal deal in writing. https://www.recordedfuture.com/research/dark-covenant-3-controlled-impunity-and-russias-cybercriminals

• Health Service Executive (Ireland) / PwC. Conti Cyber Attack on the HSE: Independent Post-Incident Review. HSE, 2021.

  The biggest attack on an Irish state agency ever, reconstructed line by line: paper records for weeks, and the anatomy of how one click took down a health service. HSE Post-Incident Review (PDF). https://www.hse.ie/eng/services/publications/conti-cyber-attack-on-the-hse-full-report.pdf (HSE: Independent Post Incident Review (current page))

• Congressional Research Service. The Change Healthcare Cyberattack and Response Considerations for Policymakers (IN12330). CRS, 2024.

  Congress’s own brief on the largest healthcare breach ever recorded: one Citrix box, no MFA, a third of US claims traffic, 192.7 million people. https://www.congress.gov/crs-product/IN12330

• Cybersecurity and Infrastructure Security Agency (CISA/FBI/HHS). Ransomware Activity Targeting the Healthcare and Public Health Sector (AA20-302A). CISA, 2020.

  The joint advisory that told hospitals the Ryuk/Conti wave was coordinated, credible, and already inside the building. https://www.cisa.gov/stopransomware/ransomware-activity-targeting-healthcare-and-public-health-sector

• Office of the Director of National Intelligence, National Counterintelligence and Security Center. China’s Collection of Genomic and Other Healthcare Data from America: Risks to Privacy and U.S. Economic and National Security. NCSC Fact Sheet, 2021.

  The counterintelligence agency stating plainly that Beijing treats bulk genomic and health data as a strategic commodity, and is collecting yours. https://www.dni.gov/files/NCSC/documents/SafeguardingOurFuture/NCSC_China_Genomics_Fact_Sheet_2021revision20210203.pdf

• UK National Cyber Security Centre (with NSA, CSE, CISA). Advisory: APT29 Targets COVID-19 Vaccine Development. NCSC, July 2020.

  Cozy Bear, three allied services, one target: the vaccine labs of the UK, US, and Canada, hit with WellMess and WellMail mid-pandemic. https://www.ncsc.gov.uk/news/advisory-apt29-targets-covid-19-vaccine-development

Content Moderation and Platform Governance

• Andrew Lowenthal (foreword by Matt Taibbi). The Censorship-Industrial Complex. Skyhorse, 2027.

  The lifelong progressive who mapped the complex for Taibbi, then wrote the book on how counterterrorism infrastructure got repurposed into ‘counter-populism.’ https://www.skyhorsepublishing.com/9781510784864/the-censorship-industrial-complex/

• Michael Shellenberger. Testimony: The Censorship-Industrial Complex. House Select Subcommittee on the Weaponization of the Federal Government, 9 March 2023.

  The man who coined the phrase, under oath, the same morning Taibbi testified. https://docs.house.gov/meetings/FD/FD00/20230309/115442/HHRG-118-FD00-Wstate-ShellenbergerM-20230309.pdf

• Lee Fang / The Intercept. Twitter Aided the Pentagon’s Covert Online Propaganda Campaign. December 2022.

  CENTCOM asking Twitter to whitelist 52 Arabic-language sock puppets ‘we use to amplify certain messages.’ https://theintercept.com/2022/12/20/twitter-dod-us-military-accounts/

• The Free Press (Bari Weiss / David Zweig). How Twitter Rigged the Covid Debate. 2022.

  Non-expert foreign contractors adjudicating American medical speech. https://www.thefp.com/p/how-twitter-rigged-the-covid-debate

• Reason. The FBI Paid Twitter $3.4 Million for Processing Requests. December 2022.

  The primary receipt for the headline number the whole FBI-Twitter section turns on. https://reason.com/2022/12/19/the-fbi-paid-twitter-3-4-million-for-processing-requests/

• CNN. No Explicit Directive: FBI and the Hunter Biden Laptop. December 2022.

  The present-all-sides counterweight: six execs and officials say no directive was given. https://www.cnn.com/2022/12/23/politics/twitter-files-elon-musk-fbi-hunter-biden-laptop/index.html

• Frances Haugen. Written Testimony, Senate Commerce Subcommittee. 5 October 2021.

  The pre-LLM template: copy the research, hand it to the Journal, testify, repeat. https://www.commerce.senate.gov/services/files/FC8A558E-824E-4914-BEDB-3A7B1190BD49

• Freedom House. Freedom on the Net 2024: China. 2024.

  Bottom of the table so reliably it’s a fixed point — ’the world’s worst conditions for internet freedom for a decade,’ in their own words. https://freedomhouse.org/country/china/freedom-net/2024

• European Union. Regulation (EU) 2022/2065 (Digital Services Act). Official Journal, 2022.

  The primary text under the Commission explainer everyone cites — systemic-risk duties and rapid-removal orders, the democratic world’s most comprehensive legal takedown machine. https://eur-lex.europa.eu/eli/reg/2022/2065/oj

• European Union. Regulation (EU) 2021/784 (Terrorist Content Online). Official Journal, 2021.

  The one-hour removal clock — a competent authority points and the platform has sixty minutes, over-removal priced in as the safe default. https://eur-lex.europa.eu/eli/reg/2021/784/oj

• United Kingdom. Online Safety Act 2023 (c. 50). legislation.gov.uk, 2023.

  The ‘highly effective age assurance’ statute whose Section 121 quietly reserves the power to order scanning of end-to-end-encrypted messages. https://www.legislation.gov.uk/ukpga/2023/50/contents

• Germany. Netzwerkdurchsetzungsgesetz (NetzDG). Gesetze im Internet, 2017.

  The primary statute behind the HRW write-up — the fine-backed 24-hour takedown template a dozen less-free governments photocopied. https://www.gesetze-im-internet.de/netzdg/

• Access Now. #KeepItOn.

  The people who count the world’s internet blackouts, and keep finding India at the top of the leaderboard year after year. https://www.accessnow.org/keepiton/

• Ministry of Electronics and IT (India). Intermediary Guidelines and Digital Media Ethics Code Rules, 2021.

  The rapid-takedown lever a functioning democracy reaches for whenever a platform hesitates. https://www.meity.gov.in/content/intermediary-guidelines-and-digital-media-ethics-code-rules-2021

• United States. 47 U.S.C. § 230. Legal Information Institute (Cornell).

  Twenty-six words that made American platforms un-sueable and, by accident, made private moderation the whole ballgame. https://www.law.cornell.edu/uscode/text/47/230

• United States. Protecting Americans from Foreign Adversary Controlled Applications Act (PAFACA), Pub. L. 118-50. 2024.

  The divest-or-die TikTok law — Washington reaching the kill switch directly, no platform intermediary required. https://www.congress.gov/118/plaws/publ50/PLAW-118publ50.pdf

• Cyberspace Administration of China. Official portal (cac.gov.cn).

  The regulator that clears speech on the way up instead of taking it down after — prior restraint with a .gov.cn address and a real-name mandate stapled to the back. http://www.cac.gov.cn/

• LiveLaw. Bombay High Court Strikes Down IT Amendment Rules 2023 Fact Check Unit. 2024.

  The check that occasionally bites: a court told the government it could not appoint itself arbiter of ‘fake’ news about the government. https://www.livelaw.in/top-stories/bombay-high-court-strikes-down-it-amendment-rules-2023-fact-check-unit-249527

• World Federation of Advertisers. WFA Discontinues GARM. August 2024.

  The advertiser cartel that policed platform ‘brand safety’ folded three days after X sued it for antitrust — the private layer dissolving the instant it was named a defendant. https://wfanet.org/knowledge/item/2024/08/09/wfa-discontinues-garm

• Global Internet Forum to Counter Terrorism. About GIFCT.

  One hash, every member platform, content gone at once — the machinery that kills a Christchurch livestream in minutes and, with the appeal step quietly missing, anything else too. https://gifct.org/about/

• Guillaume Chaslot / AlgoTransparency. AlgoTransparency. 2016–present.

  The ex-YouTube engineer who built a lie detector for the recommendation engine that fired him. It kept flagging flat earth. https://www.algotransparency.org/

• Homa Hosseinmardi et al. Causally Estimating the Effect of YouTube’s Recommender System on Watch Behavior. PNAS, 2024.

  The rabbit hole is real but narrow — ~3% of users, and the algorithm mostly follows the appetite rather than manufacturing it. https://www.pnas.org/doi/10.1073/pnas.2213020120

• Ferenc Huszár et al. Algorithmic Amplification of Politics on Twitter. PNAS 119(1), 2022.

  Twitter’s own data: mainstream right amplified more than mainstream left in six of seven countries — and not the fringe. https://www.pnas.org/doi/10.1073/pnas.2025334119

• Soroush Vosoughi, Deb Roy, Sinan Aral. The Spread of True and False News Online. Science 359(6380), 2018.

  126,000 stories, falsehood winning every category. The lie was faster because it was newer, and the humans did the spreading. https://www.science.org/doi/10.1126/science.aap9559

• Jeff Horwitz. Broken Code: Inside Facebook and the Fight to Expose Its Harmful Secrets. Doubleday, 2023.

  The reporter Haugen handed the cache to, showing his work at book length. The company documented the harm in PowerPoint, then filed it. https://www.penguinrandomhouse.com/books/670585/broken-code-by-jeff-horwitz/

  • Amazon: https://www.amazon.com/dp/0385549180

• Eli Pariser. The Filter Bubble: What the Internet Is Hiding from You. Penguin Press, 2011.

  The thesis everyone cites and the literature later half-refuted. Kept for the argument, not the verdict. https://en.wikipedia.org/wiki/The_Filter_Bubble

  • Amazon: https://www.amazon.com/dp/1594203008

• Christopher A. Bail et al. Exposure to Opposing Views on Social Media Can Increase Political Polarization. PNAS 115(37), 2018.

  Show a Republican a month of liberal tweets and he comes out more conservative. The cure made the disease worse. https://www.pnas.org/doi/10.1073/pnas.1804840115

• Bureau of Investigative Journalism. London PR Firm Rewrites Wikipedia for Governments and Billionaires. January 2026.

  Eight years of covert edits — Qatar’s World Cup, a Libyan sovereign-wealth faction — through a sockpuppet ring that got caught. The expose is also the alibi: the system that busted it works. https://www.thebureauinvestigates.com/stories/2026-01-14/london-pr-firm-rewrites-wikipedia-for-governments-and-billionaires

• Atlantic Council. Katherine Maher on How Big Tech Can Be as Trusted as Wikipedia (transcript). June 2021.

  The former Wikimedia CEO on the record: ‘a very active approach to disinformation,’ a clearinghouse built ’through conversations with government, of course.’ https://www.atlanticcouncil.org/news/transcripts/katherine-maher-on-how-big-tech-can-be-as-trusted-as-wikipedia/

• Times of Israel. Edit Wars Over Israel Spur Rare Ban of 8 Wikipedia Editors — From Both Sides. 2025.

  ArbCom’s own machinery found coordinated editing and sanctioned six pro-Palestine and two pro-Israel editors. The both-sides fact everyone selling a purge ignores. https://www.timesofisrael.com/edit-wars-over-israel-spur-rare-ban-of-8-wikipedia-editors-from-both-sides/

• Ashley Rindsberg / The Algemeiner. Wikipedia’s Quiet Revolution. December 2024.

  The ‘Gang of 40’ accusation, attributed to its author and no further — the claim that needed ArbCom to corroborate the mechanism it points at. https://www.algemeiner.com/2024/12/04/wikipedias-quiet-revolution-how-coordinated-group-editors-reshaped-israeli-palestinian-narrative/

• Supreme Court of the United States. Moody v. NetChoice, LLC (No. 22-277). July 2024.

  Nine-nil vacate-and-remand: the Court sent Texas HB20 and Florida SB7072 back for the First Amendment homework the circuits skipped. https://www.supremecourt.gov/opinions/23pdf/22-277_d18f.pdf

• EFF. Visa and Mastercard Are Trying to Dictate What You Can Watch on Pornhub. December 2020.

  The card networks discover they are a content-moderation department, and the EFF points out nobody elected them to it. https://www.eff.org/deeplinks/2020/12/visa-and-mastercard-are-trying-dictate-what-you-can-watch-pornhub

• ADL. Bad Gateway: How Deplatforming Affects Extremist Websites. 2023.

  The DNS/registrar layer, measured: kick a site off the stack and its traffic craters. The empirical spine under the Daily Stormer / 8chan anecdotes. https://www.adl.org/resources/report/bad-gateway-how-deplatforming-affects-extremist-websites

• NPR. Meet the Man Behind Epik, the Tech Firm Keeping Far-Right Websites Alive. February 2021.

  ‘The registrar of last resort,’ self-billed as the Swiss bank of domains — the one door left when every other layer shuts. https://www.npr.org/2021/02/08/965448572/meet-the-man-behind-epik-the-tech-firm-keeping-far-right-websites-alive

• EFF. The Internet Is Not Facebook: Why Infrastructure Providers Should Stay Out of Content Policing. October 2022.

  The distinction the whole deplatforming fight turns on, argued by the group that later had to revise itself over Kiwi Farms. https://www.eff.org/deeplinks/2022/10/internet-not-facebook-why-infrastructure-providers-should-stay-out-content

• FIRE. Statement on Free Speech and Online Payment Processors.

  A First Amendment shop putting the payment rail on the free-speech map, where the money is the choke and the choke is the point. https://www.fire.org/research-learn/fire-statement-free-speech-and-online-payment-processors

• NPR. Judge Refuses to Reinstate Parler After Amazon Shut It Down. January 2021.

  ‘Dwindlingly slight,’ the judge called the antitrust claim. One company hosts a third of the internet and gets to say no. https://www.npr.org/2021/01/21/956486352/judge-refuses-to-reinstate-parler-after-amazon-shut-it-down

• Tarleton Gillespie. Content Moderation, AI, and the Question of Scale. Big Data & Society 7(2), 2020.

  The line the whole scale problem hangs on: “It is size, not scale, that makes automation seem necessary.” A 99%-accurate filter run across billions of posts is wrong millions of times, and who it’s wrong about is never random. https://doi.org/10.1177/2053951720943234

• Harvard Journal of Law & Technology. The Meta Oversight Board and the Empty Promise of Legitimacy. Vol. 37 No. 2.

  The board that overturns Meta most of the time and changes nothing structural. Legitimacy theater with a very high reversal rate. https://jolt.law.harvard.edu/assets/articlePDFs/v372/3-The-Meta-Oversight-Board-and-the-Empty-Promise-of-Legitimacy.pdf

• Jacob Mchangama and Joelle Fiss. The Digital Berlin Wall: How Germany (Accidentally) Created a Prototype for Global Online Censorship. Justitia, 2019 (Act 2, 2020).

  Germany built a takedown machine to fight hate speech; the Danish think tank counted the regimes that xeroxed it. First edition: at least 13 countries. The 2020 update: 25, most of them “not free.” The template travels better than the intent. https://globalfreedomofexpression.columbia.edu/publications/the-digital-berlin-wall-how-germany-accidentally-created-a-prototype-for-global-online-censorship/

• Tech Transparency Project. Apple Is Censoring its App Store for China. 2021.

  3,257 apps gone from the China store, 964 of them on privacy tools, Tibetan Buddhism, Hong Kong, LGBTQ. The company that runs the “what happens on your iPhone stays on your iPhone” ad, editing the shelf to Beijing’s spec. https://www.techtransparencyproject.org/articles/apple-censoring-its-app-store-china

• 404 Media / Yahoo News. “Wikipedia Co-Founder Larry Sanger Banned From the Site.” June 2026.

  English Wikipedia indefinitely banned its own co-founder on a “canvassing” technicality after he launched a WikiProject to counter what he called the site’s bias against currently disfavoured views and groups. Co-founder Jimmy Wales publicly criticized the ban. The reference institution purged the dissenter on procedure, not substance — the gatekeeper defending itself. 404 Media; Yahoo News; Slashdot

• Larry Sanger. “I Co-Founded Wikipedia. Then an Anonymous Mob Banned Me.” Washington Examiner, July 8, 2026.

  The founder’s own account, and a clean three-move capture on a reference institution. Move 1 — take the institution: Sanger describes control passing to “a few hundred anonymous demagogues,” a “relatively small clique of anonymous administrators” wielding “irresistible authority” — permeation by the admin corps, no cabal required. Move 2 — control the language: the machinery runs on “vague rules, vague threats, and vague alliances,” so a charge like “not here to build the encyclopedia” means whatever the clique needs it to mean; the label does the work argument used to. Move 3 — the ratchet: the co-founder is permanently blocked from the thing he co-founded, three days of abuse first, the door closing behind him. He also names the interested-party layer the June coverage didn’t: journalist Ashley Rindsberg’s “Gang of 40” accounts dominating Israel–Palestine coverage; former CEO Katherine Maher’s 2021 Atlantic Council remarks about Wikipedia’s “active approach to disinformation” in coordination with governments (the statecraft-wing tell); the Bureau of Investigative Journalism’s January 2026 finding that London PR firm Portland Communications rewrote articles for clients including Qatar, the Gates Foundation, and Libya’s sovereign wealth fund; and WikiScanner’s 2006 exposure of edits from CIA IP ranges. All claims above are Sanger’s, cited as his; the point for the series is structural — “the free encyclopedia anyone can edit” is now one of the most powerful media platforms in history, governed by people no one can name. Washington Examiner

• Matt Taibbi et al. “The Twitter Files.” Racket News / Substack, December 2022 – March 2023.

  19 installments. FBI paying Twitter $3.4 million. Secret blacklists. CENTCOM sock puppets. The Virality Project flagging true stories. https://twitterfiles.substack.com/

• Meta Oversight Board. 2023 Annual Report. 2024.

  Overturned Meta’s original decision ~90% of the time. The body designed to legitimize the system kept finding the system wrong. https://www.oversightboard.com/news/2023-annual-report-shows-boards-impact-on-meta/

• Human Rights Watch. “Meta’s Broken Promises: Systemic Censorship of Palestine Content.” December 2023.

  Arabic hostile speech classifier over-removes Palestinian speech while under-removing Hebrew-language incitement. https://www.hrw.org/report/2023/12/21/metas-broken-promises/systemic-censorship-palestine-content-instagram-and

• EFF. “Unfiltered: How YouTube’s Content ID Discourages Fair Use.” 2016.

  2.2 billion claims in 2024. 65% of disputes resolved in the uploader’s favor — meaning the system is wrong more often than right when challenged. https://www.eff.org/wp/unfiltered-how-youtubes-content-id-discourages-fair-use-and-dictates-what-we-see-online

• Human Rights Watch. “Germany: Flawed Social Media Law.” February 2018.

  NetzDG: the template. 24-hour takedown deadlines and EUR 50 million fines. At least 13 countries copied it, including Russia and Venezuela. https://www.hrw.org/news/2018/02/14/germany-flawed-social-media-law

• House Judiciary Committee. “Staff Report on the Election Integrity Partnership Jira Tickets.” November 2023.

  The pipeline: government flags content to Stanford → Stanford flags to platforms → platforms act. Separation maintained at every stage. https://judiciary.house.gov/sites/evo-subsites/republicans-judiciary.house.gov/files/evo-media-document/EIP_Jira_Ticket_Staff_Report_11-6-23_Clean.pdf

• EFF. “Beyond Platforms: Private Censorship, Parler, and the Stack.” January 2021.

  The essay that identified the infrastructure stack as a chain of chokepoints. “There is no way to separate the infrastructure provider from the content it hosts.” https://www.eff.org/deeplinks/2021/01/beyond-platforms-private-censorship-parler-and-stack

• CNBC. “Cloudflare CEO Admits That Removing Neo-Nazi Site Because He’s in a ‘Bad Mood’ Is a Slippery Slope.” August 2017.

  Matthew Prince: “I woke up in a bad mood and decided someone shouldn’t be allowed on the Internet. No one should have that power.” Then he did it again. And again. https://www.cnbc.com/2017/08/17/cloudflare-ceo-says-removing-the-daily-stormer-is-slippery-slope.html

• TechCrunch. “Cloudflare Will Stop Service to 8chan.” August 2019.

  After Christchurch and El Paso. Prince called 8chan “a cesspool of hate.” The exception became the precedent. https://techcrunch.com/2019/08/04/cloudflare-will-stop-service-to-8chan-which-ceo-matthew-prince-describes-as-a-cesspool-of-hate/

• Cloudflare Blog. “Blocking Kiwifarms.” September 2022.

  “An unprecedented emergency and immediate threat to human life.” The threshold lowers each time. https://blog.cloudflare.com/kiwifarms-blocked/

• House Judiciary Committee. “The Weaponization of CISA.” June 2023 (PDF).

  CISA’s MDM team coordinated directly with platforms. Facebook created a special government portal for CISA to flag content. https://judiciary.house.gov/sites/evo-subsites/republicans-judiciary.house.gov/files/evo-media-document/cisa-staff-report6-26-23.pdf

• House Judiciary Committee. “Final Report on the Weaponization of the Federal Government.” 2024.

  17,000 pages. 99 closed-door deposition transcripts. The receipts. https://judiciary.house.gov/media/press-releases/final-report-weaponization-federal-government

• The Intercept. “Leaked Documents Outline DHS’s Plans to Police Disinformation.” October 2022.

  Internal DHS memos: monitoring speech on COVID origins, vaccine efficacy, racial justice, Afghanistan withdrawal, and Ukraine. https://theintercept.com/2022/10/31/social-media-disinformation-dhs/

• House Judiciary Committee. “NSF Is Funding AI Tools to Censor.” 2023.

  $38.8 million in NSF grants for “misinformation” research. 64 grants to 42 universities. https://judiciary.house.gov/media/press-releases/new-report-details-how-national-science-foundation-funding-ai-tools-censor-and

• Matt Taibbi. Congressional testimony on the Censorship-Industrial Complex. March 2023 (PDF).

  Named ~20 quasi-private entities scanning social media alongside government agencies. https://docs.house.gov/meetings/FD/FD00/20230309/115442/HHRG-118-FD00-Wstate-TaibbiM-20230309.pdf

• Reason. “Twitter Files Show Employees Knew Russian Bots List Was Fake.” January 2023.

  Hamilton 68: the flagged accounts were overwhelmingly real people (~86% English-language), not strongly Russian or bot, per Taibbi’s Twitter Files 15; the Washington Post issued corrections. Hundreds of media stories cited it anyway. https://reason.com/2023/01/27/twitter-files-matt-taibbi-hamilton-68-russian-bots-fake/

• Reason. “State Department Funds a Disinformation Index That Rates Conservative Outlets as ‘Risky.’” February 2023.

  GDI received $330,000 from the State Department while rating Reason, the NYP, and RealClearPolitics as “high risk.” https://reason.com/2023/02/14/global-disinformation-index-state-department-list-risk-reason/

• Evelyn Douek. “Content Moderation as Systems Thinking.” Harvard Law Review, 2022.

  The most sophisticated academic defense of platform governance as systemic infrastructure. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4005326

• 7amleh — The Arab Center for the Advancement of Social Media. Erased and Suppressed. 2024.

  Twenty testimonies from Palestinian journalists on how Meta’s Arabic “hostile speech” classifier scrubbed their reach. The field receipts behind the “over-removes Palestinian, under-removes Hebrew” finding. https://7amleh.org/storage/meta/Erased%20and%20Suppressed%20-1812.pdf

• EFF. We Need to Talk About Infrastructure. EFF, 2022.

  EFF’s follow-on to “The Internet Is Not Facebook”: conceding that what counts as essential infrastructure keeps moving, which is exactly how every chokepoint gets reclassified as fair game. https://www.eff.org/deeplinks/2022/12/we-need-talk-about-infrastructure

• ARTICLE 19. Internet: Content Moderation at Infrastructure Level Puts Rights at Risk. ARTICLE 19, 2021.

  The free-expression NGO’s warning that pushing takedowns down to the DNS and registrar layer is a blunt instrument with no due process: the rights case under the Daily Stormer and 8chan anecdotes. https://www.article19.org/resources/icann-content-moderation-at-the-infrastructure-level-is-a-dangerous-move/

• Andrew M. Guess et al. How Do Social Media Feed Algorithms Affect Attitudes and Behavior in an Election Campaign? Science 381(6656):398-404, 2023.

  The other half of the amplification story: switch consenting users to a chronological feed for three months and their politics barely move. The algorithm shapes the diet, not necessarily the voter. https://www.science.org/doi/10.1126/science.abp9364

• Killian L. McLoughlin, William J. Brady, et al. Misinformation Exploits Outrage to Spread Online. Science 386(6725):991-996, 2024.

  A million Facebook links and the mechanism drops out clean: false news evokes more outrage, outrage gets shared unread, and nobody had to check whether it was true. https://www.science.org/doi/10.1126/science.adl2829

• C. Thi Nguyen. Echo Chambers and Epistemic Bubbles. Episteme 17(2):141-161, 2020.

  The distinction the whole debate needed: a bubble just leaves voices out and pops on contact with evidence; an echo chamber actively discredits the outside, so exposure only tightens it. https://doi.org/10.1017/epi.2018.32

• Amy Ross Arguedas, Craig T. Robertson, Richard Fletcher, and Rasmus Kleis Nielsen. Echo Chambers, Filter Bubbles, and Polarisation: A Literature Review. Reuters Institute for the Study of Journalism, 2022.

  The receipt for the deflation: echo chambers rarer than advertised, no support for the filter-bubble hypothesis, most people on a boringly centrist media diet. https://reutersinstitute.politics.ox.ac.uk/echo-chambers-filter-bubbles-and-polarisation-literature-review

• ADL Center for Technology and Society. Exposure to Alternative and Extremist Content on YouTube. Anti-Defamation League, 2021.

  The counterweight to the rabbit-hole panic: extremist exposure is real but concentrated in viewers who already carry the resentment, and it arrives mostly by subscription and off-platform links, not the recommender. https://www.adl.org/resources/report/exposure-alternative-extremist-content-youtube

• Information Technology and Innovation Foundation. The UK’s Online Safety Act’s Predictable Consequences Are a Cautionary Tale for the US. ITIF, September 2025.

  A DC tech think tank scoring the OSA’s first year: 1,400% VPN spikes, enforcement theater, and the age-check-as-ID-onramp the sponsors swore it wasn’t. https://itif.org/publications/2025/09/03/the-uks-online-safety-acts-predictable-consequences-are-cautionary-tale-for-the-us/

• Sophia Cope, Aaron Mackey, and Andrew Crocker. The EARN IT Act Violates the Constitution. Electronic Frontier Foundation, March 2020.

  The legal case that conditioning Section 230 on undefined best practices is a warrantless-search machine aimed at end-to-end encryption: the Barr-era backdoor bill by another name. https://www.eff.org/deeplinks/2020/03/earn-it-act-violates-constitution

• Christophe Geiger and Bernd Justin Jütte. Platform Liability Under Article 17 of the Copyright in the Digital Single Market Directive, Automated Filtering and Fundamental Rights: An Impossible Match. GRUR International 70(6), 2021.

  Two copyright scholars do the arithmetic the EU legislature skipped: you cannot mandate upload filters and honor fundamental rights at once. The title is the finding. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3776267

• Spandana Singh. Everything in Moderation: An Analysis of How Internet Platforms Are Using Artificial Intelligence to Moderate User-Generated Content. New America Open Technology Institute, 2019.

  The pre-LLM survey that catalogued the failure modes early, dataset bias and no context and no transparency, then watched every platform scale the tools anyway. https://www.newamerica.org/oti/reports/everything-moderation-analysis-how-internet-platforms-are-using-artificial-intelligence-moderate-user-generated-content/

• Maarten Sap, Dallas Card, Saadia Gabriel, Yejin Choi, and Noah A. Smith. The Risk of Racial Bias in Hate Speech Detection. Proceedings of ACL 2019.

  The primary under the AAVE-bias headline: tweets in Black English up to twice as likely to be tagged offensive, the bias baked into the training data and dutifully reproduced by the model. https://aclanthology.org/P19-1163/

• Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu, and James Zou. GPT Detectors Are Biased Against Non-Native English Writers. Patterns (Cell Press) 4(7), 2023.

  Over half of TOEFL essays flagged as machine-written while native eighth-graders sail through: the detector mistakes a smaller vocabulary for a robot, and the accusation lands on the foreign student. https://www.cell.com/patterns/fulltext/S2666-3899(23)00130-7

• U.S. House Committee on Oversight and Accountability. Hearing Document: Protecting Speech from Government Interference and Social Media Bias (Twitter Files exhibits). 8 February 2023.

  The visibility-filtering receipts entered into the congressional record: Trends Blacklisted, Search Blacklisted, and a company insisting under its own 2018 words that it didn’t shadow ban. https://www.congress.gov/118/meeting/house/115286/documents/HHRG-118-GO00-20230208-SD004.pdf

• 51 Former Senior Intelligence Officials. Public Statement on the Hunter Biden Emails. Politico, 19 October 2020.

  Fifty-one ex-spooks signing that the laptop had all the classic earmarks of a Russian information operation; organizer Morell later testified under oath the intent was a debate talking point, not an assessment. https://www.politico.com/f/?id=00000175-4393-d7aa-af77-579f9b330000

• African Content Moderators Union. Founding vote, Nairobi. 1 May 2023.

  More than 150 Facebook, TikTok and ChatGPT moderators and data labelers voted to form the first content moderators’ union on the continent. The organising tier, not the labour-market statistics: this is the antibody response, cited in The Last Antibodies ch17. https://time.com/6275995/chatgpt-facebook-african-workers-union/

• Foxglove. The Nairobi content moderation summit, where the union vote took place. 15 May 2023.

  First-hand account of the summit and the vote. Foxglove is co-counsel in the Kenyan moderator litigation, so read as an interested party with direct knowledge. https://www.foxglove.org.uk/2023/05/15/nairobi-content-moderation-summit/

• TIME. Kauna Malgwi — the 100 Most Influential People in AI. 2024.

  Nigerian psychology graduate, joined Sama’s Nairobi hub in 2019 moderating Hausa-language content, chairs the union’s Nigeria chapel, and is among the former moderators bringing the action against Meta and Sama. Listed alongside the founders whose products she was cleaning. https://time.com/collections/time100-ai-2024/7012799/kauna-malgwi/

• Kenya, Employment and Labour Relations Court (Ongaya, J.). Ruling that Meta is the primary employer of the content-moderation work. 2 June 2023.

  THE receipt for the ch17 argument: no new AI statute was needed. Ordinary employment law held that the party setting the terms is the employer, and it reached Meta from a Nairobi courtroom. The court also suspended the mass layoff. Meta had argued it was not a proper party because it is not registered in Kenya. Under appeal; a holding, not a final judgment. https://startupnews.fyi/2023/06/03/kenyan-court-rules-meta-as-primary-employer-in-content-moderators-lawsuit/

• Business & Human Rights Resource Centre. Court of Appeal rules Meta can be sued in Kenyan courts over content-moderator layoffs.

  Appellate confirmation of Kenyan jurisdiction over Meta. Tracks the procedural history and the unpaid redundancy payments the moderators say they are owed. https://www.business-humanrights.org/en/latest-news/court-of-appeal-rules-meta-can-be-sued-in-kenyan-courts-over-layoffs-of-content-moderators/

• openDemocracy. Meta can be sued by ex-Facebook moderator, Kenyan court rules.

  Daniel Motaung’s May 2022 suit against Meta and Samasource Kenya alleging exploitation, union-busting and pay discrimination. PENDING — the claims are allegations and are carried as allegations in the prose. https://www.opendemocracy.net/en/5050/meta-lawsuit-kenya-facebook-content-moderator-sued-daniel-motaung/

Disinformation and Deepfakes

• OECD AI Incidents Monitor. “Hungarian PM Posts AI-Generated Deepfake Video, Prompting EU Warning.” 6 February 2026.

  A sitting prime minister posts a fake of the Commission president with his challenger; 1.5 million views; the Commission asks him nicely. https://oecd.ai/en/incidents/2026-02-06-9252

• OpenAI. An Update on Disrupting Deceptive Uses of AI (the ‘Zero Zeno’ takedown). OpenAI, 30 May 2024.

  The first time OpenAI publicly named and removed a state-linked op, and it named Israel in the same report as Russia, China, and Iran. The built-in anti-exceptionalism anchor. https://openai.com/global-affairs/an-update-on-disrupting-deceptive-uses-of-ai/

• The Forward. US Pro-Israel Groups Failed to Disclose Grants From Israeli Government (Concert / Kela Shlomo). 2021.

  Ministry money structured so recipients ‘see a bank transfer from a non-profit,’ none filed under FARA: the disclosure-gap receipt. https://forward.com/israel/453286/us-pro-israel-groups-failed-to-disclose-grants-from-israeli-government/

• Mozilla Foundation. YouTube Regrets (RegretsReporter Final Report). 2021.

  The crowdsourced study everyone cites and no one reads; 71% of regrets came from YouTube’s own recommendations. https://www.mozillafoundation.org/en/youtube/findings/

• Georgetown CSET. “Truth, Lies, and Automation: How Language Models Could Change Disinformation.” 2023.

  Human-curated AI propaganda more persuasive than human-written alone: 53% vs 47%. https://cset.georgetown.edu/

• Imperva. “2025 Bad Bot Report.” 2025.

  Bots now 51% of all web traffic. The dead internet theory, confirmed in the aggregate. https://www.imperva.com/resources/resource-library/reports/2025-bad-bot-report/

• NewsGuard. “AI-Generated News Websites.” 2024.

  3,749 AI-generated news and information sites across 16 languages as of the current tracker (up from 2,089), still adding hundreds a month. https://www.newsguardtech.com/special-reports/ai-tracking-center/

• Yale ISPS. “The Liar’s Dividend.” 2024.

  Politicians who dismiss real scandals as deepfakes gain support. The better deepfakes get, the easier it becomes to deny reality. https://isps.yale.edu/research/publications/isps24-07

• CNN. “Finance Worker Pays Out $25 Million After Video Call with Deepfake CFO.” May 2024.

  Arup engineering firm. Every participant on the video call was a deepfake. $25 million in 15 transfers. https://www.cnn.com/2024/05/16/tech/arup-deepfake-scam-loss-hong-kong-intl-hnk

• Surfshark. “Deepfake Election Incidents.” 2024.

  38 countries, 3.8 billion people affected since 2021. Romania annulled a presidential election — first in European history. https://surfshark.com/research/chart/election-related-deepfakes

• NPR. “FCC AI Deepfake Robocall Biden.” May 2024.

  Created in under 20 minutes for $1. $6M FCC fine, 26 criminal counts. https://www.npr.org/2024/05/23/nx-s1-4977582/fcc-ai-deepfake-robocall-biden-new-hampshire-political-operative

• Microsoft. “As the U.S. Election Nears, Russia, Iran, and China Step Up Influence Efforts.” October 2024.

  Russia’s Doppelganger spoofs news sites. China’s Spamouflage. Iran’s STORM-2035. State-sponsored disinformation at industrial scale. https://blogs.microsoft.com/on-the-issues/2024/10/23/as-the-u-s-election-nears-russia-iran-and-china-step-up-influence-efforts/

• Georgetown CSET. “How Persuasive Is AI-Generated Propaganda?” PNAS Nexus, February 2024.

  Human-curated AI propaganda: 53% agreement. Human-written alone: 47%. AI-only: 43%. Human-machine teaming wins. https://cset.georgetown.edu/publication/how-persuasive-is-ai-generated-propaganda/

• Euronews. “Spanish Teens Received Deepfake AI Nudes of Themselves.” September 2023.

  28 girls aged 11-17 in Almendralejo. AI nudes generated from Instagram posts. Distributed by classmates. https://www.euronews.com/next/2023/09/24/spanish-teens-received-deepfake-ai-nudes-of-themselves-but-is-it-a-crime

• Nature. “AI-Generated Papers and the Academic Fraud Problem.” 2025.

  Hindawi retracted 8,000+ articles. The rat with giant genitals. 60,000+ papers with undisclosed AI use. https://www.nature.com/articles/d41586-025-02906-y

• FBI. “Public Service Announcement on Generative AI Fraud.” December 2024.

  AI-enabled scams up 456% year-over-year. 82% of phishing emails now AI-assisted. https://www.ic3.gov/PSA/2024/PSA241203

• ABC News. “Mom Warns of Hoax AI Clone of Daughter’s Voice.” April 2023.

  Jennifer DeStefano: “My daughter’s voice, 100%.” AI cloned a 15-year-old’s voice for a $1M kidnapping scam. https://abcnews.go.com/GMA/Family/mom-warns-hoax-ai-clone-daughters-voice/story?id=98551351

• ADL. “The Safety Divide: Open-Source AI Models Fall Short on Guardrails.” 2025.

  HiddenLayer bypassed guardrails using “emoji smuggling.” Open-source without guardrails generates content closed models won’t. https://www.adl.org/resources/report/safety-divide-open-source-ai-models-fall-short-guardrails-antisemitic-dangerous

• Reco AI. “AI and Cloud Security Breaches 2025.” 2025.

  Chinese state actors used Claude Code to automate intrusions against ~30 organizations. 4-6 human decision points per campaign. https://www.reco.ai/blog/ai-and-cloud-security-breaches-2025

• Eric Goldman. “ChatGPT Defeats Defamation Lawsuit Over Hallucination — Walters v. OpenAI.” Technology & Marketing Law Blog, May 2025.

  The first AI-defamation suit, lost: a Georgia court held ChatGPT’s fabricated embezzlement claim about a named radio host wasn’t defamatory because no reasonable reader believes the machine. The defense is that nobody should have believed it. https://blog.ericgoldman.org/archives/2025/05/chatgpt-defeats-defamation-lawsuit-over-hallucination-walters-v-openai.htm

• Cleary Gottlieb. “Georgia Court Dismisses Defamation Lawsuit Against OpenAI Over ChatGPT Output.” 2025.

  The doctrinal write-up of the Walters v. OpenAI summary judgment — the legal logic of treating a generative model as a known fabricator. https://www.clearygottlieb.com/news-and-insights/publication-listing/georgia-court-dismisses-defamation-lawsuit-against-openai-over-chatgpt-output

• Renée DiResta and Josh A. Goldstein. “How Spammers and Scammers Leverage AI-Generated Images on Facebook for Audience Growth.” arXiv:2403.12838, 2024.

  The Shrimp Jesus economy, documented: the slop is not ideological, it is just cheaper than hiring anyone, and the Feed serves it to people who never followed the page. https://arxiv.org/abs/2403.12838

• Jonas Dolezal, Sawood Alam, Mark Graham, Maty Bohacek. “The Impact of AI-Generated Text on the Internet.” arXiv:2604.26965, 2026.

  By mid-2025 roughly 35% of newly published sites read as AI-made, up from ~zero pre-ChatGPT. But when they tested six Dead-Internet fears against the data, only two survived: shrinking diversity and a drift toward the artificially cheerful. The evidence says flattening, not rot. https://arxiv.org/abs/2604.26965

• Originality.ai. AI Content in Google Search Results. 2025 (ongoing tracker).

  A detector vendor grading the top-20 Google results: 2.27% AI in Feb 2019, a 19.56% peak in July 2025, settling to 17.31% by September. High-ranking results, measured by a firm with a commercial stake, and it says so if you read the fine print. https://originality.ai/ai-content-in-google-search-results

• Kaitlyn Tiffany. “The ‘Dead-Internet Theory’ Is Wrong but Feels True.” The Atlantic, August 31, 2021.

  The piece that carried a chan-board metaphysics into the mainstream by taking it seriously instead of laughing (“wrong but feels true”), and in doing so handed the theory the oxygen its own believers never could. https://www.theatlantic.com/technology/archive/2021/08/dead-internet-theory-wrong-but-feels-true/619937/

• Gopani, Avi. “Conspiracy Theorists Says The Internet Has Been Dead Since 2016.” Analytics India Magazine, 6 September 2021.

  The genealogy of record: the January 2021 Agora Road post, the Wizardchan and 4chan antecedents, and the observable-core-vs-strong-psyop split, documented on the one platform now measurably filling with the machine text the theory was about.

  • URL: https://analyticsindiamag.com/conspiracy-theorists-says-the-internet-has-been-dead-since-2016/
  • Wikipedia (index): https://en.wikipedia.org/wiki/Dead_Internet_theory

• Kai-Cheng Yang and Filippo Menczer. “Anatomy of an AI-Powered Malicious Social Botnet.” arXiv:2307.16336, 2023.

  The astroturf prediction, automated and caught in the wild: ~1,140 X accounts running ChatGPT, and the AI-text detectors, OpenAI’s own among them, could not reliably flag it. The authors call it the tip of the iceberg, which is the optimistic reading. https://arxiv.org/abs/2307.16336

• arXiv. “The Rise of AI-Generated Content in Wikipedia.” arXiv:2410.08044, 2024.

  The flood reaching the reference layer itself, measured on the encyclopedia the corpus already documents as a capture target. https://arxiv.org/abs/2410.08044

• Originality.ai. Over Half of Long Posts on LinkedIn Are Likely AI-Generated Since ChatGPT Launched. 2024.

  A detector vendor grading the platform’s long-form output and finding most of it machine-made post-ChatGPT. A detector estimate with an uncharacterized false-positive rate, which is exactly how to read it. https://originality.ai/blog/ai-content-published-linkedin

• Originality.ai. 15% of Reddit Posts Are Likely AI-Generated in 2025. 2025.

  Same vendor, the forum layer: about 15% for 2025. A detector estimate, not a count. https://originality.ai/blog/ai-reddit-posts-study

• Kate Knibbs. “AI Slop Is Flooding Medium.” WIRED, October 29, 2024.

  The commissioned sample behind the 274,466-post / 47% figure, and the story where Medium’s CEO called the detector firm’s follow-up sales pitch “extortion by press,” a small monument to the era. https://www.wired.com/story/ai-generated-medium-posts-content-moderation/

• Futurism. “Experts: 90% of Online Content Will Be AI-Generated by 2026.” 2022.

  The scary round number: an expert estimate from a Europol Innovation Lab report, popularized past its evidence. Cite as a forecast, never as a measurement. https://futurism.com/the-byte/experts-90-online-content-ai-generated

• Tech Business News. “The Problem With Europol’s 90% AI Content Prediction: Numbers Don’t Add Up.” 2023.

  The debunk, filed beside the claim so the projection never travels alone. https://www.techbusinessnews.com.au/blog/the-problem-with-europols-90-ai-content-prediction-numbers-dont-add-up/

• NPR. “AI Books Are Crowding the Marketplace on Amazon.” March 13, 2024.

  The book layer going the way of the blog layer: Amazon capping self-published uploads at three a day and mandating AI disclosure, measures reported as useless against the volume. https://www.npr.org/2024/03/13/1237888126/growing-number-ai-scam-books-amazon

Financial Control and CBDCs

• India Stack. Open APIs (Aadhaar, UPI, Account Aggregator). iSPIRT.

  The export catalog for the identity-and-payments stack: the open-source rail that ships abroad as ‘digital public infrastructure,’ off-switch and all. https://indiastack.org/

• Federal Reserve. FedNow Service — About.

  The Fed’s instant-settlement plumbing, and its own insistence that it ‘is not related to a digital currency.’ https://www.federalreserve.gov/paymentsystems/fednow_about.htm

• Federal Reserve Board. Federal Reserve Board Announces Withdrawal from the NGFS. 2025.

  The Fed quietly walked out of the green-central-bank book club three days before the inauguration — the pawl called permanent, unclicking itself. https://www.federalreserve.gov/newsevents/pressreleases/bcreg20250117a.htm

• ESG Today. Net Zero Banking Alliance Ceases Operations. 2025.

  $74T alliance to zero members: the banking pillar didn’t lose six members, it turned off the lights. https://www.esgtoday.com/net-zero-banking-alliance-ceases-operations/

• Sabin Center for Climate Change Law. SEC Decides to End Legal Defense of its Climate Disclosure Rules. 2025.

  The disclosure mandate the book called a locked click — the SEC stopped defending its own rule in court. https://climate.law.columbia.edu/content/sec-decides-end-legal-defense-its-climate-disclosure-rules

• Accountancy Europe. Omnibus Explained: Key Changes to the CSRD and CSDDD. 2025.

  Brussels descoped ~42,000 firms and slid the deadline two years — ‘mandatory for 50,000 firms’ got renegotiated in daylight. https://accountancyeurope.eu/publications/omnibus-explained-key-changes-to-the-csrd-and-csddd/

• Net Zero Asset Managers initiative. Statement on BlackRock’s Departure from the Initiative. 2025.

  NZAM pulled its own signatory list off the web pending ‘review’ the week BlackRock left — the asset-manager pillar suspending itself. https://www.netzeroassetmanagers.org/statement-on-blackrocks-departure-from-the-initiative/

• U.S. House Committee on Financial Services. Operation Chokepoint 2.0: Biden’s Debanking of Digital Assets. November 2025.

  The receipt that upgrades ‘Choke Point 2.0’ from crypto-Twitter grievance to congressional finding: 30+ digital-asset entities debanked by informal pressure, no rule anyone could sue over. https://financialservices.house.gov/uploadedfiles/2025-11-30_–_fsc_debanking_report_final_1.pdf

• CoinDesk. U.S. Regulator Told Banks to Lay Off Crypto, Letters Obtained by Coinbase Reveal. December 2024.

  The FOIA’d ‘pause all crypto-asset activity’ letters, in the FDIC’s own hand — Choke Point 2.0 with a paper trail after all. https://www.coindesk.com/policy/2024/12/05/u-s-regulator-told-banks-to-lay-off-crypto-letters-obtained-by-coinbase-reveal/

• The Financial Brand. Chase Cancels Kanye West — But Should Banks Engage in Culture Wars? 2022.

  The trade press for bankers asking, out loud, whether the debanking trigger should be commercial toxicity — the industry watching itself become the switch. https://thefinancialbrand.com/news/banking-trends-strategies/chase-cancels-kanye-west-but-should-banks-engage-in-culture-wars-154537

• FDIC Office of Inspector General. The FDIC’s Role in Operation Choke Point (Report AUD-15-008). September 2015.

  The regulator’s own watchdog on Choke Point — and the sting in the tail: it called the FDIC’s role ‘inconsequential’ and found no targeting, only that officials ‘did not always follow’ policy. https://www.fdicoig.gov/sites/default/files/reports/2022-08/15-008AUD.pdf

• European Central Bank. Digital euro — FAQ.

  The promise on the box: ‘does not support a programmable digital euro that would restrict how users can spend their money.’ A stated intention is not a tested constraint. https://www.ecb.europa.eu/euro/digital_euro/faqs/html/ecb.faq_digital_euro.en.html

• Nicholas Anthony. Digital Currency or Digital Control? Decoding CBDC and the Future of Money. Cato Institute, 2024.

  The book-length case that a US CBDC trades privacy, financial freedom, and central-bank independence for benefits the private market already delivers. Cato’s monetary analyst — the person who actually counted the Fed’s comment letters. https://www.cato.org/books/digital-currency-or-digital-control

  • Amazon: https://www.amazon.com/dp/1964524385

• RUSI. Tom Keatinge. Down the Rabbit Hole: Central Bank Digital Conspiracies. Commentary, 26 April 2024.

  The steel-man from the security-studies side — separates the documented surveillance concern from the goldbug conspiracy, which is exactly why it belongs in a sourced dossier. https://www.rusi.org/explore-our-research/publications/commentary/down-rabbit-hole-central-bank-digital-conspiracies

• BIS. Agustin Carstens. “Digital Currencies and the Soul of Money.” January 2022.

  “The central bank will have absolute control on the rules and regulations that will determine the use of that expression of central bank liability.” https://www.bis.org/speeches/sp220118.htm

• CNAS. Yaya Fanusie and Emily Jin. “China’s Digital Currency: Adding Financial Data to Digital Authoritarianism.” January 2021.

  “DCEP technical architecture will allow for real-time or near-real-time financial surveillance of all users’ transactions.” https://s3.us-east-1.amazonaws.com/files.cnas.org/documents/CNAS-Report-Chinas-Digital-Currency-Jan-2021-final.pdf

• CNBC. “How Parler’s Deplatforming Shows the Power of Cloud Providers.” January 2021.

  Google, Apple, then AWS — in three days. The full infrastructure cascade. https://www.cnbc.com/2021/01/16/how-parler-deplatforming-shows-power-of-cloud-providers.html

• Reuters. “FBI Finds Limited Evidence That Jan. 6 Capitol Breach Was Coordinated.” August 20, 2021.

  FBI’s own investigation found the Capitol breach was not centrally coordinated. Facebook groups — not Parler — were the primary organizational venue. https://www.reuters.com/world/us/exclusive-fbi-finds-scant-evidence-us-capitol-attack-was-coordinated-sources-2021-08-20/

• ProPublica. “Facebook Hosted Surge of Misinformation and Insurrection Threats in Months Leading Up to Jan. 6 Attack, Records Show.” January 2022.

  Internal Facebook documents showed the platform’s own algorithms amplified Stop the Steal groups to millions. Parler caught the killshot. Facebook kept its app store listings. https://www.propublica.org/article/facebook-hosted-surge-of-misinformation-and-insurrection-threats-in-months-leading-up-to-jan-6-attack-records-show

• U.S. Senate Committee on Homeland Security and Governmental Affairs. “Examining the January 6 Attack on the U.S. Capitol.” 2021.

  Senate investigation documented Facebook as the primary platform where Jan 6 events were organized and promoted. https://www.hsgac.senate.gov/media/majority-media/peters-klobuchar-release-joint-report-on-january-6th-attack-on-the-u-s-capitol/

• The Conversation. “How Authorities Target the Convoy Money via the Emergencies Act.” February 2022.

  Banks froze accounts without court orders. Federal Court later ruled it unconstitutional. https://theconversation.com/how-authorities-are-targeting-the-freedom-convoy-money-via-the-emergencies-act-177204

• WikiLeaks. “Banking Blockade.” 2010.

  Bank of America, Visa, Mastercard, PayPal, Western Union destroyed 95% of WikiLeaks revenue. No criminal charges filed. https://wikileaks.org/Banking-Blockade.html

• TMZ. “Kanye West’s Bank JPMorgan Chase Cuts Ties with Yeezy Brand.” October 2022.

  JPMorgan sent the termination letter September 20 — before the most inflammatory statements. The banking kill switch was already engaged. https://www.tmz.com/2022/10/12/kanye-jp-morgan-chase-end-banking-relationship-yeezy-candace-owens/

• The Business Standard. “Kanye West Melts Down on Twitter, Claims Kim Kardashian Tried to Lock Him Up.” 2020.

  Ye publicly alleged that Kim’s family tried to have him institutionalized — flying a doctor to Wyoming “to lock me up.” The man screaming about handlers was screaming because there were handlers. (His own posts, Feb–Oct 2022, are the primary record below.) https://www.tbsnews.net/glitz/kanye-west-melts-down-twitter-claims-kim-kardashian-tried-lock-him-109561

• Ye (Kanye West). Instagram posts, February–October 2022.

  Months of public posts naming specific lawyers, managers, and handlers he said were controlling his access to his children. Deleted by platform. Widely screenshotted and archived.

• NBC News. “Kanye West and Kim Kardashian’s Custody Battle: What We Know.” November 2022.

  Joint custody granted, but Ye was publicly fighting over school choice, residency, and access — while the people managing both sides of the dispute were also managing his career. https://www.nbcnews.com/pop-culture/pop-culture-news/kanye-west-kim-kardashian-custody-agreement-divorce-rcna59221

• Fortune. “These Brands and Companies That Have Dropped Kanye.” October 26, 2022.

  The cascade in real time. Adidas, Balenciaga, Gap, CAA, Foot Locker, TJ Maxx — every layer of commercial infrastructure, simultaneously. https://fortune.com/2022/10/26/kanye-west-ye-brands-companies-dropped-adidas-balenciaga-gap-caa/

• CBS News. “Kanye West Is No Longer a Billionaire, Forbes Says.” October 2022.

  Forbes tracked the collapse: ~$2 billion to ~$400 million. The Adidas partnership alone was $1.5 billion of the loss. https://www.cbsnews.com/news/kanye-west-list-of-companies-dropped-ye-no-longer-bilionaire-antisemitic-comments/

• Al Jazeera. “Adidas Ends Partnership with Kanye West Over Antisemitic Remarks.” October 25, 2022.

  $246 million in Q4 losses. $1.3 billion in unsold Yeezy inventory. Ten percent of Adidas revenue, terminated in a press release. https://www.aljazeera.com/economy/2022/10/25/adidas-ends-partnership-with-kanye-west-over-antisemetic-remarks

• CNBC. “NatWest CEO Alison Rose Resigns After Media Storm.” July 26, 2023.

  Nigel Farage’s bank compiled a 40-page dossier on his political views. CEO denied it. Dossier leaked. CEO resigned. https://www.cnbc.com/2023/07/26/nigel-farage-coutts-saga-natwest-ceo-alison-rose-resigns-after-media-storm.html

• Nic Carter. “Operation Choke Point 2.0.” February 2023.

  FDIC told banks to “pause all crypto-asset activity.” FOIA’d letters to two dozen banks. 30+ digital asset companies de-banked. https://cryptoslate.com/is-the-crypto-industry-in-danger-vc-nic-carter-says-operation-choke-point-2-0-is-well-underway/

• Marc Andreessen on The Joe Rogan Experience (#2234). November 26, 2024.

  “I know of thirty founders who have been de-banked.” The venture capitalist who built Netscape, documenting infrastructure-level financial exclusion. https://open.spotify.com/episode/0U4Y0an6iKM1VB8pE1M4vQ

• Federal Court of Canada. Canadian Civil Liberties Association v. Canada (Attorney General). January 23, 2024.

  Justice Mosley: Emergencies Act invocation was unreasonable. The government froze 257 accounts in hours. The court said it was unconstitutional two years later. https://www.cato.org/blog/emergencies-act-after-two-years

• Federal Court of Appeal (Canada). Canada v. Canadian Civil Liberties Association. January 16, 2026.

  Upheld. The invocation was unreasonable, ultra vires, and infringed Charter rights. Government appealing to SCC. https://www.cbc.ca/news/politics/convoy-protest-emergencies-act-appeal-9.7046769

• House Oversight Committee. “Staff Report: Operation Choke Point.” May 2014.

  Obama DOJ pressured banks into dropping legal-but-disfavored industries. Payday lenders, gun dealers, tobacconists. https://oversight.house.gov/wp-content/uploads/2014/05/Staff-Report-Operation-Choke-Point1.pdf

• IMF. Bo Li. “CBDC and Programmability.” October 2022.

  “By programming CBDC, those money can be precisely targeted for what kind of people can own and what kind of use this money can be utilized.” https://www.imf.org/en/news/seminars/campaigns/2022/10/annual-meetings-day5-2

• White House. “EO 14178: Strengthening American Leadership in Digital Financial Technology.” January 2025.

  Prohibits federal agencies from establishing CBDCs. The US killed its CBDC — on paper. https://www.whitehouse.gov/presidential-actions/2025/01/strengthening-american-leadership-in-digital-financial-technology/

• Carnegie Endowment. “Why Nigeria’s Naira Redesign Policy Hasn’t Met Its Objectives.” January 2024.

  Old notes pulled faster than new distributed. Lines outside banks turned to protests turned to riots. https://carnegieendowment.org/research/2024/01/why-nigerias-controversial-naira-redesign-policy-hasnt-met-its-objectives

• Financial Conduct Authority. FCA Sets Out Initial Findings on Bank Account Access and Closures. FCA, 2023.

  The regulator’s own count after the Farage storm: ~343,000 accounts terminated in a year, and no firm evidence anyone was closed for their politics; dormancy and financial-crime concerns did the work. The receipt that keeps the debanking panic honest. https://www.fca.org.uk/news/press-releases/fca-sets-out-initial-findings-bank-account-access-and-closures

• Abdul Latif Jameel Poverty Action Lab (J-PAL). Digital Financial Services to Improve Formalized Access and Inclusion. J-PAL policy insight.

  The steel-man for the ID rail: RCT evidence that digital financial services widen formal inclusion, the benefit the exclusion deaths are weighed against. https://www.povertyactionlab.org/policy-insight/digital-financial-services-improve-formalized-access-and-inclusion

• Atlantic Council. What to Watch as China Prepares Its Digital Yuan for Prime Time. Econographics, 2026.

  The think-tank scorecard on the largest live CBDC: 3.4 billion transactions, 16.7 trillion yuan, and a January reclassification from digital cash to deposit money that nobody voted on. https://www.atlanticcouncil.org/blogs/econographics/what-to-watch-as-china-prepares-its-digital-yuan-for-prime-time/

• Carnegie Endowment for International Peace. What Will Be the Impact of China’s State-Sponsored Digital Currency? Carnegie, 2021.

  Early Carnegie read on the e-CNY: the surveillance ledger arrives dressed as convenience, and the export version is the point. https://carnegieendowment.org/posts/2021/07/what-will-be-the-impact-of-chinas-state-sponsored-digital-currency

• Lawfare. The Programmable State: The e-CNY and China’s Quest for Smarter Surveillance. Lawfare, 2026.

  Programmability isn’t a bug the Party tolerates, it’s the feature: money that enforces its own conditions, from subsidy rails to expiry dates. https://www.lawfaremedia.org/article/the-programmable-state–the-e-cny-and-china-s-quest-for-smarter-surveillance

• U.S. Congress. H.R. 1919: Anti-CBDC Surveillance State Act. 119th Congress, 2025.

  The statutory no to a digital dollar: passed the House 219-210 on party lines, barring the Fed from issuing a retail CBDC. The surveillance objection, made law. https://www.congress.gov/bill/119th-congress/house-bill/1919

• Michelle W. Bowman (Federal Reserve). Considerations for a Central Bank Digital Currency. Speech, 18 April 2023.

  A sitting Fed governor on the record: a CBDC could intrude on the privacy of users and harm the banking system while providing few benefits. The house skeptic, in the house’s own transcript. https://www.federalreserve.gov/newsevents/speech/bowman20230418a.htm

• Christopher J. Waller (Federal Reserve). CBDC: A Solution in Search of a Problem? Speech, 5 August 2021.

  The title is the argument: highly skeptical, what problem would a CBDC solve? From the governor who said the quiet part into a microphone. https://www.federalreserve.gov/newsevents/speech/waller20210805a.htm

• Nicholas Anthony. CBDC Spells Doom for Financial Privacy. Cato Institute, Fall 2024.

  Cato’s monetary analyst, article-length: a CBDC offers neither cash’s privacy nor its finality, and puts the government at the center of every transaction. Companion to his book already in this list. https://www.cato.org/free-society/fall-2024/cbdc-spells-doom-financial-privacy

• Brandon Joel Tan. Central Bank Digital Currency and Financial Inclusion. IMF Working Paper WP/23/69, 2023.

  The steel-man from inside the Fund: in developing economies a CBDC could bank the unbanked, and, in the same breath, sells the credit-scoring data-from-CBDC-usage that turns the wallet into a file. https://www.imf.org/en/publications/wp/issues/2023/03/18/central-bank-digital-currency-and-financial-inclusion-531104

• Jookyung Ree. Nigeria’s eNaira, One Year After. IMF Working Paper WP/23/104, 2023.

  The primary behind the failure number: the IMF’s own count that 98.5% of eNaira wallets were never used once. Africa’s first CBDC, audited and found empty. https://www.imf.org/en/publications/wp/issues/2023/05/16/nigerias-enaira-one-year-after-533487

Infrastructure and Deplatforming Law

• Ganesh Sitaraman. “Deplatforming.” Yale Law Journal, Vol. 133, No. 2. November 2023.

  The legal framework the chapter hangs on. Duty-to-serve vs. right-to-exclude. Neither extreme applies; the internet’s infrastructure layer has no framework at all. https://yalelawjournal.org/pdf/133.2.Sitaraman_fmyqyznb.pdf

• Columbia Global Freedom of Expression. Parler LLC v. Amazon Web Services, Inc. 2021.

  Court found coordination evidence “dwindlingly slight.” Didn’t address whether infrastructure-level deplatforming should require any process at all. https://globalfreedomofexpression.columbia.edu/cases/parler-v-amazon-web-services/

Education and Student Surveillance

• ACLU. Digital Dystopia: The Danger in Buying What the EdTech Surveillance Industry Is Selling. 2023.

  A $3.1-billion-a-year ’economic juggernaut’ whose safety claims the ACLU calls ‘unsubstantiated and unverifiable.’ An industry that sells peace of mind and measures nothing. https://www.aclu.org/publications/digital-dystopia-the-danger-in-buying-what-the-edtech-surveillance-industry-is-selling

• Center for Democracy & Technology. Off Task: EdTech Threats to Student Privacy and Equity in the Age of AI. 2024.

  Near-universal deployment (89% of teachers), monitoring used mostly to discipline, and 60% of students who won’t type what they think while it’s watching. The chilling effect, quantified. https://cdt.org/insights/report-off-task-edtech-threats-to-student-privacy-and-equity-in-the-age-of-ai/

• EFF (Jason Kelley). ExamSoft Flags One-Third of California Bar Exam Test Takers for Cheating. December 2020.

  A cheating detector so trigger-happy it accuses a third of the future bar. Lenovo users caught en masse by microphone failures. https://www.eff.org/deeplinks/2020/12/examsoft-flags-one-third-california-bar-exam-test-takers-cheating

• EFF. EFF Client Erik Johnson and Proctorio Settle Lawsuit Over Bogus DMCA Claims. March 2022.

  A student reverse-engineered Proctorio, tweeted the receipts, and got hit with DMCA takedowns for his trouble. He sued and won. https://www.eff.org/deeplinks/2022/03/eff-client-eric-johnson-and-proctorio-settle-lawsuit-over-bogus-dmca-claims

• EFF. How GoGuardian Invades Student Privacy. October 2023.

  Over twenty-seven million students across ten thousand schools. Monitoring that follows the school Chromebook home, onto the kid’s own WiFi, after hours. https://www.eff.org/deeplinks/2023/10/how-goguardian-invades-student-privacy

• BuzzFeed News (Caroline Haskins). Gaggle Knows Everything About Teens And Kids In School. 2019.

  An algorithm that reads every Google Doc, email, and chat a student writes on a school account, and routes the flags to humans who call the school. https://www.buzzfeednews.com/article/carolinehaskins1/gaggle-school-surveillance-technology-education

• The 74 (Mark Keierleber). Gaggle Drops LGBTQ Keywords From Student Surveillance Tool Following Bias Concerns. 2023.

  ‘Gay,’ ’lesbian,’ and ’transgender’ filed under ‘sexual content’ until January 2023. The Trevor Project returned the donation. https://www.the74million.org/article/gaggle-drops-lgbtq-keywords-from-student-surveillance-tool-following-bias-concerns/

• The 74 (Mark Keierleber). Survey Reveals Extent That Cops Surveil Students Online — in School and at Home. 2023.

  89% of teachers, software used punitively, law enforcement ’literally into kids’ homes.’ https://www.the74million.org/article/survey-reveals-extent-that-cops-surveil-students-online-in-school-and-at-home/

• Supreme Court of the United States. Mahanoy Area School District v. B.L. 594 U.S. ___ (2021).

  8-1: a school can’t punish a cheerleader’s off-campus weekend Snapchat. Breyer worried about schools regulating ‘all the speech a student utters’ — in a world where the school laptop already logs it. https://www.supremecourt.gov/opinions/20pdf/20-255_g3bi.pdf

• Danielle Keats Citron. The Surveilled Student. 76 Stanford Law Review 1439, 2024.

  The law-review case that continuous school monitoring is itself the injury, not merely a prelude to punishment. https://review.law.stanford.edu/wp-content/uploads/sites/3/2024/10/Citron-76-Stan.-L.-Rev.-1439.pdf

• EPIC. In re Online Test Proctoring Companies. Complaint and Request for Investigation, Federal Trade Commission, December 2020.

  The formal complaint that named the whole racket at once, Respondus and ProctorU and Proctorio and Examity and Honorlock, for unfair and deceptive practices, excessive biometric collection, and opaque, biased AI. https://epic.org/documents/in-re-online-test-proctoring-companies/

• Center for Democracy & Technology. Commercial Companies and FERPA’s School Official Exception: A Survey of Privacy Policies. CDT, 2024.

  How a 1974 law meant to protect student records became the loophole EdTech vendors drive through: access without parental consent, no security floor, no vendor liability. https://cdt.org/insights/commercial-companies-and-ferpas-school-official-exception-a-survey-of-privacy-policies/

• Center for Democracy & Technology. The Chilling Effect of Student Monitoring: Disproportionate Impacts and Mental Health Risks. CDT, 2023.

  Sixty percent of students won’t type what they actually think when the software is watching, and the ones flagged hardest are poor, disabled, or LGBTQ. The self-censorship is the product working as designed. https://cdt.org/insights/the-chilling-effect-of-student-monitoring-disproportionate-impacts-and-mental-health-risks/

• Privacy International. Studying Under Surveillance: The Securitisation of Learning. Privacy International long-read, 2023.

  The classroom reframed as a threat surface, where learning is something to be monitored, secured, and flagged rather than done. https://privacyinternational.org/long-read/5463/studying-under-surveillance-securitisation-learning

• Harvard Law Review. Mahanoy Area School District v. B. L. Vol. 135, 2021.

  The law-review reading of the cheerleader case: the Court drew a line at off-campus speech but left the school laptop that logs it all firmly inside the fence. https://harvardlawreview.org/print/vol-135/mahanoy-v-b-l/

• ACLU. Cops and No Counselors: How the Lack of School Mental Health Staff Is Harming Students. ACLU, 2019.

  The report behind the numbers: schools with police arrest students at 3.5 times the rate of those without, up to eight times in some states, mostly for nonviolent offenses. Millions of kids have a cop but no counselor. https://www.aclu.org/publications/cops-and-no-counselors

AI Chatbot Liability, Litigation and Regulation

• Foundation for Individual Rights and Expression. Order on Motion to Dismiss — Garcia v. Character Technologies, Inc. M.D. Fla., 2025.

  The ruling that told an industry its chatbot is a power tool, not a poem: a product for strict liability, not speech for the First Amendment. Section 230 wasn’t even attempted. https://www.fire.org/research-learn/order-motion-dismiss-garcia-v-character-technologies-inc

• Sen. Steve Padilla (CA). First-in-Nation AI Chatbot Safeguards Signed Into Law. 2025.

  SB 243: the machine now has to admit it isn’t human, and remind the minor every three hours in case they forgot. Effective January 1, 2026. https://sd18.senate.ca.gov/news/first-nation-ai-chatbot-safeguards-signed-law

• Future of Privacy Forum. Understanding the New Wave of Chatbot Legislation: California SB 243 and Beyond. 2025.

  The neutral map of who’s regulating companion bots and how — the sober counterweight to the press-release triumphalism. https://fpf.org/blog/understanding-the-new-wave-of-chatbot-legislation-california-sb-243-and-beyond/

• Morrison Foerster. New York and California Enact Landmark AI Companion Laws. 2025.

  Two states, six weeks apart, and a $15,000-a-day New York penalty clock — the first real price tag on an unlabeled companion bot. https://www.mofo.com/resources/insights/251120-new-york-and-california-enact-landmark-ai

• Matthew Raine. Written Testimony, U.S. Senate Committee on the Judiciary. September 2025.

  A father telling the Senate what his sixteen-year-old’s chatbot said back. The primary document under the Raine v. OpenAI headlines. https://www.judiciary.senate.gov/imo/media/doc/e2e8fc50-a9ac-05ec-edd7-277cb0afcdf2/2025-09-16%20PM%20-%20Testimony%20-%20Raine.pdf

• Social Media Victims Law Center & Tech Justice Law Project. Lawsuits Accuse ChatGPT of Emotional Manipulation and Acting as a Suicide Coach. November 2025.

  Seven suits in one filing, four of them wrongful-death, all built on the same claim: the sycophancy was the product, not a bug. https://socialmediavictims.org/press-releases/smvlc-tech-justice-law-project-lawsuits-accuse-chatgpt-of-emotional-manipulation-supercharging-ai-delusions-and-acting-as-a-suicide-coach/

• Utah Division of Consumer Protection. Utah v. Snap, Inc. — Complaint. June 2025.

  The state’s own managers reportedly called My AI’s rollout ‘reckless.’ A chatbot you can’t delete, coaching minors on where to hide the liquor. https://commerce.utah.gov/wp-content/uploads/2025/06/2025.06.30-Public-UT-Snap-Complaint.pdf

• Kentucky Office of the Attorney General. Kentucky v. Character Technologies — Complaint. January 2026.

  First state AG to sue Character.AI directly, and the one that noticed the children’s chat logs quietly became training data. https://www.ag.ky.gov/Press%20Release%20Attachments/CTI%20Complaint%20Motion%20and%20Order%20Filed.pdf

• Gibson Dunn. Gibson Dunn Wins Significant Victory for OpenAI (Walters v. OpenAI). 2025.

  The defense bar’s victory lap on Walters — a win welded so tightly to its facts it may not survive the next plaintiff who actually lost a job. https://www.gibsondunn.com/gibson-dunn-wins-significant-victory-for-client-openai-defending-against-defamation-claim-based-on-hallucinated-generative-ai-output/

• McGuireWoods. Can Social Media or AI Be a Defective Product? 2026.

  The doctrinal spine: the Social Media MDL’s ‘functionality-based’ product test that Garcia leaned on to keep Section 230 out of the room. https://www.mcguirewoods.com/client-resources/alerts/2026/3/can-social-media-or-ai-be-a-defective-product/

• Moody’s. Section 230 Immunity for AI Chatbot Lawsuits. 2025.

  Why the shield built for hosting other people’s speech goes limp when the machine is the one doing the talking. https://www.moodys.com/web/en/us/insights/insurance/230-immunity-for-AI-chatbot-lawsuits.html

• 119th Congress. TAKE IT DOWN Act (S.146). U.S. Congress, 2025.

  The one AI bill that actually became law, and it’s about deepfake nudes, not the companion bot that talks a kid into the noose. Congress fixed the photogenic harm. https://www.congress.gov/bill/119th-congress/senate-bill/146

• 119th Congress. Kids Online Safety Act (KOSA, S.1748). U.S. Congress, 2025.

  Passed the Senate 91-3, died in the House, came back as S.1748, still isn’t law. The child-safety bill everyone votes for and nobody enacts. https://www.congress.gov/bill/119th-congress/senate-bill/1748/text

• 119th Congress. GUARD Act: Guidelines for User Age-verification and Responsible Dialogue Act (S.3062). U.S. Congress, 2025.

  Government ID at the door before a minor can talk to a companion bot, and a federal crime for the bot that coaches self-harm. Advanced by committee, not yet law. https://www.congress.gov/bill/119th-congress/senate-bill/3062/text

• 119th Congress. NO FAKES Act (H.R.2794). U.S. Congress, 2025.

  Hollywood’s bill against digital replicas of its stars: the celebrity likeness gets statutory protection the suicidal teenager never got. https://www.congress.gov/bill/119th-congress/house-bill/2794/text

• California State Legislature. SB 243: Companion Chatbots. 2025.

  The statute behind the Padilla press release: admit you’re a machine, re-admit it every three hours to the minor, publish a crisis protocol, or pay $1,000 a violation. https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202520260SB243

• Skadden, Arps, Slate, Meagher & Flom. The TAKE IT DOWN Act. Skadden, 2025.

  The defense bar’s explainer on the sole enacted AI statute: notice-and-takedown by May 2026, and nothing whatsoever for companion-bot harm. https://www.skadden.com/insights/publications/2025/06/take-it-down-act

Legal Cases and Regulation

• ProPublica / The Lever. The Man Behind the Right’s Takeover of the Federal Courts (Barre Seid to Marble Freedom Trust). August 2022.

  The $1.6B that turned a Chicago electronics fortune into Leonard Leo’s judicial-selection war chest, disclosed by tax records, not by anyone volunteering. https://www.propublica.org/article/dark-money-leonard-leo-barre-seid

• Jack Goldsmith. The Terror Presidency: Law and Judgment Inside the Bush Administration. W. W. Norton, 2007.

  The OLC head who withdrew Yoo’s memos, then wrote the book about what the office does under pressure.

  • Amazon: https://www.amazon.com/dp/039333533X

• U.S. Department of Justice, Office of the Inspector General. Review of Four FISA Applications and Other Aspects of the FBI’s Crossfire Hurricane Investigation. December 2019.

  Horowitz on the 17 FISA errors: the primary receipt for the FISA-process critique. https://oig.justice.gov/reports/review-four-fisa-applications-and-other-aspects-fbis-crossfire-hurricane-investigation

• CSIS (Gregory Allen et al.). Understanding the Biden Administration’s Updated Export Controls. CSIS, 2023.

  The map of the hardware Great Firewall, drawn by the think tank that helped survey the route. https://www.csis.org/analysis/understanding-biden-administrations-updated-export-controls

• U.S. Bureau of Industry and Security. Commerce Announces Rescission of Biden-Era AI Diffusion Rule. BIS, May 13, 2025.

  The government revoking its own most sweeping chip rule the week before it bit — proof the ratchet occasionally slips when Nvidia’s lobbyists pull hard enough. https://www.bis.gov/press-release/department-commerce-announces-rescission-biden-era-artificial-intelligence-diffusion-rule-strengthens

• CNBC. Nvidia Says U.S. Government Will Allow It to Resume H20 AI Chip Sales to China. July 15, 2025.

  Export control quietly reborn as a 15%-of-revenue toll booth: the leash became a tollgate. https://www.cnbc.com/2025/07/15/nvidia-says-us-government-will-allow-it-to-resume-h20-ai-chip-sales-to-china.html

• NIST. AI Risk Management Framework (AI RMF 1.0), NIST AI 100-1. 2023.

  GOVERN/MAP/MEASURE/MANAGE — the voluntary grammar the apparatus adopts as its own permits. NIST — AI Risk Management Framework 1.0 (PDF). https://nvlpubs.nist.gov/nistpubs/ai/nist.ai.100-1.pdf

• GovTrack. H.Amdt. 155 (Massie) to H.R. 7148 — vote. 2026.

  The ‘kill switch’ amendment, voted down so the mandate keeps its funding. Democracy declining to un-ratchet itself, on the record. https://www.govtrack.us/congress/votes/119-2026/h43

• Snopes. Did Congress Pass a Law Requiring a ‘Kill Switch’ in Cars?

  The fact-check that says ’no remote kill switch’ and accidentally documents the capability being built anyway. https://www.snopes.com/fact-check/congress-cars-kill-switch-law/

• DADSS Program. Driver Alcohol Detection System for Safety — Program Overview.

  The sensor that decides whether your breath lets the car start. Licensable to OEMs ‘by end of 2025,’ allegedly. https://dadss.org/

• MADD. 10 Things to Know About the Drunk Driving Prevention Technology Provision.

  The org that buried its children behind this technology also wrote down, in advance, that it could become a privacy weapon. https://madd.org/press-release/10-things-to-know-about-the-drunk-driving-prevention-technology-provision-in-the-infrastructure-bill/

• CNBC. U.S. to Investigate National Security Data Risks From Chinese Vehicles. February 2024.

  A Commerce Secretary worrying out loud that foreign modules could ‘disable vehicles remotely’ — confirming the capability is real by fearing the wrong hand on it. https://www.cnbc.com/2024/02/29/us-says-to-investigate-national-security-data-risks-from-chinese-vehicles.html

• ETSC. Intelligent Speed Assistance (ISA).

  The safety case is airtight; the ‘overridable until it isn’t’ firmware toggle is the part nobody votes on. https://etsc.eu/intelligent-speed-assistance-isa/

• European Road Safety Charter. Intelligent Speed Assistance Set to Become Mandatory Across Europe.

  Every new EU car since July 2024 can already read and enforce the limit. The pedal resists; one update makes it refuse. https://road-safety-charter.ec.europa.eu/resources-knowledge/media-and-press/intelligent-speed-assistance-isa-set-become-mandatory-across

• Supreme Court. Murthy v. Missouri. 23-411, June 26, 2024.

  6-3 on standing. Punted on the merits. The question of whether government jawboning violates the First Amendment remains unanswered. https://www.supremecourt.gov/opinions/23pdf/23-411_3dq3.pdf

• Supreme Court. Murthy v. Missouri. 23-411, oral argument transcript, March 18, 2024.

  https://www.supremecourt.gov/oral_arguments/argument_transcripts/2023/23-411_5367.pdf

• ProPublica. “Machine Bias: There’s Software Used Across the Country to Predict Future Criminals. And It’s Biased Against Blacks.” May 2016.

  COMPAS: Black defendants twice as likely to be mislabeled as high-risk. The article that launched a thousand dissertations. https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing

• SBS. “Robodebt Royal Commission Report.” 2023.

  433,000 unlawful debts. 663 people died. Rhys Cauzzo received 12 letters and made 13 calls before he died at 28. https://www.sbs.com.au/news/article/criminal-charges-cruel-treatment-the-main-takeaways-from-the-robodebt-report/5jphhfnw6

• European Commission. “The Digital Services Act.” 2024.

  165 million content moderation decisions appealed. 30% (almost 50 million) reversed. 14 investigations opened. https://digital-strategy.ec.europa.eu/en/policies/digital-services-act

• ACLU. “Senate Reauthorizes and Expands Section 702 Surveillance.” April 2024.

  Expanded the definition of “electronic communications service provider” so broadly that critics said any business with Wi-Fi could be compelled to assist NSA surveillance. https://www.aclu.org/press-releases/senate-reauthorizes-and-expands-section-702-surveillance

• EFF. “Salt Typhoon Hack Shows There’s No Security Backdoor That’s Only for the ‘Good Guys.’” October 2024.

  They warned about CALEA in 1994. Thirty years later: “The lesson will be repeated until it is learned.” https://www.eff.org/deeplinks/2024/10/salt-typhoon-hack-shows-theres-no-security-backdoor-thats-only-good-guys

• Anzalone Law. “Jury Finds Tesla Autopilot Defective in Landmark $329 Million Verdict.” September 2025.

  First jury finding Tesla Autopilot defective. 67% driver fault, 33% Tesla. https://www.anzalonelaw.com/jury-finds-tesla-autopilot-defective-in-landmark-329-million-verdict-what-it-means-for-future-cases/

• White House. “Executive Order: Eliminating State Law Obstruction of National AI Policy.” December 2025.

  States with “onerous” AI laws lose federal broadband funds. The government weaponizing infrastructure funding against AI regulation. https://www.whitehouse.gov/presidential-actions/2025/12/eliminating-state-law-obstruction-of-national-artificial-intelligence-policy/

• EU AI Act. “Implementation Timeline.” 2024.

  Only 3 of 27 member states designated enforcement authorities. Full implementation delayed to 2027. https://artificialintelligenceact.eu/implementation-timeline/

• IAPP. “Global AI Governance: China.” 2026.

  No comprehensive AI law yet. 30+ new standards expected. Incremental, standards-driven approach. https://iapp.org/resources/article/global-ai-governance-china/

• CIO Dive. “Trump Rebrands US AI Safety Institute to CAISI.” 2026.

  Removed “safety” from the name. Nearly 500 NIST employees fired or laid off. https://www.ciodive.com/news/Trump-US-AI-Safety-Institute-CAISI-rebrand/749988/

• Whistleblowers Blog. “OpenAI Whistleblowers File Complaint with SEC on Illegal NDAs.” July 2024.

  SEC complaint alleging four violations. Additional sourcing for OpenAI NDA controversy. https://whistleblowersblog.org/corporate-whistleblowers/sec-whistleblowers/openai-whistleblowers-file-complaint-with-sec-on-illegal-ndas/

• Civil Rights Litigation Clearinghouse. Mobley v. Workday, Inc. No. 3:23-cv-00770 (N.D. Cal.).

  The case that moved liability from the employer who bought the screener to the vendor who built it. Judge Lin: Workday is an “agent.” 1.1 billion applications, Workday’s own count. https://clearinghouse.net/case/44074/

• U.S. District Court, N.D. Cal. Mobley v. Workday — Order on Motion to Dismiss. July 12, 2024.

  The agent holding in the court’s own words: a software vendor can be directly liable for discriminatory screening under Title VII, ADEA, and ADA. https://www.govinfo.gov/content/pkg/USCOURTS-cand-3_23-cv-00770/pdf/USCOURTS-cand-3_23-cv-00770-1.pdf

• U.S. Equal Employment Opportunity Commission. iTutorGroup to Pay $365,000 to Settle EEOC Discriminatory Hiring Suit. August 2023.

  The software was told to auto-reject women 55+ and men 60+. Discovered only because one applicant filed twice with two birth dates and got the interview on the younger one. https://www.eeoc.gov/newsroom/itutorgroup-pay-365000-settle-eeoc-discriminatory-hiring-suit

• EEOC. Select Issues: Assessing Adverse Impact in Software, Algorithms, and AI Used in Employment Selection Procedures Under Title VII. May 18, 2023 (withdrawn from eeoc.gov; Wayback).

  The guidance that held employers liable for a vendor’s disparate impact even when the vendor certifies “bias-free.” The current EEOC quietly took it down; the archive kept the receipt. http://web.archive.org/web/20250125163154/https://www.eeoc.gov/laws/guidance/select-issues-assessing-adverse-impact-software-algorithms-and-artificial

• NYC Department of Consumer and Worker Protection. Automated Employment Decision Tools (Local Law 144). Effective July 2023.

  Annual independent bias audits, published, with 10-business-day candidate notice. Penalties $500-$1,500 per day. https://www.nyc.gov/site/dca/about/automated-employment-decision-tools.page

• Illinois General Assembly. Public Act 103-0804 (HB 3773). Signed August 9, 2024; effective January 1, 2026.

  Amends the Human Rights Act to bar discriminatory AI in hiring and firing, and bans ZIP code as a proxy for a protected class. https://www.ilga.gov/legislation/PublicActs/View/103-0804

• Colorado General Assembly. SB 24-205, Consumer Protections for Artificial Intelligence. Signed May 17, 2024.

  A “consequential decisions” framework for employment, housing, credit, education, and health, then delayed as the compliance cost sank in. https://leg.colorado.gov/bills/sb24-205

• HR Brew / Norton Rose Fulbright. Kistler et al. v. Eightfold AI, Inc. (Contra Costa County Superior Court, filed January 20, 2026).

  Filed by former EEOC chair Jenny Yang and Towards Justice. Not a discrimination suit, contrary to early framing: an FCRA/ICRAA consumer-report theory that the screener compiles undisclosed applicant reports and scores 0-5. https://natlawreview.com/article/ai-hiring-under-fire-what-eightfold-lawsuit-means-every-employer-using-algorithmic

• U.S. Congress. Reforming Intelligence and Securing America Act (RISAA), H.R. 7888, 118th Congress. 2024.

  The reauthorization that renamed the surveillance and called it reform: passed at midnight the night 702 lapsed, House 273-147, Senate 60-34, and it widened the definition of provider until a Wi-Fi router counts as a conscript. https://www.congress.gov/bill/118th-congress/house-bill/7888

• D. Adam Candeub. Amicus Curiae Brief of the “Twitter Files” Journalists, Murthy v. Missouri (No. 23-411). Supreme Court of the United States, 8 February 2024.

  The reporters who broke the files filing the mechanism straight to the Court: government points, private intermediary shoots, and Bantam Books keeps everyone’s hands clean. https://www.supremecourt.gov/DocketPDF/23/23-411/300197/20240208220913623_44722%20pdf%20Candeub.pdf

• European Commission (AI Office). The General-Purpose AI Code of Practice. European Commission, 2025.

  A voluntary code nearly 1,000 stakeholders helped draft, that regulators will treat as proof of compliance: the soft-law instrument doing hard-law work, exactly as advertised. https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai

• United States. Executive Order 14110: Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence. Federal Register, 2023.

  Biden’s reporting thresholds and red-team disclosure rules: the order Trump revoked while keeping the scaffolding it stood on. https://www.federalregister.gov/documents/2023/11/01/2023-24283/safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence

• United States. Executive Order 14179: Removing Barriers to American Leadership in Artificial Intelligence. Federal Register, 2025.

  The revocation. It struck the order and left NIST, the AI Safety Institute, and the export controls standing: proof of what a single administration can and cannot unwind. https://www.federalregister.gov/documents/2025/01/31/2025-02172/removing-barriers-to-american-leadership-in-artificial-intelligence

• Gavin Newsom. SB 1047 Veto Message. Office of the Governor of California, 29 September 2024.

  Praised the intent, killed the bill: the veto the it-rolled-back camp cites, and the incumbents quietly wanted. https://www.gov.ca.gov/wp-content/uploads/2024/09/SB-1047-Veto-Message.pdf

• United States. CHIPS and Science Act of 2022 (H.R. 4346). U.S. Congress, 2022.

  $52.7B in subsidies with a ten-year leash: take the money, don’t expand advanced fabs in China. Industrial policy as a chokepoint you sign up for. https://www.congress.gov/bill/117th-congress/house-bill/4346

• Special Counsel John H. Durham. Report on Matters Related to Intelligence Activities and Investigations Arising Out of the 2016 Presidential Campaigns. U.S. Department of Justice, May 2023.

  Four years of work to conclude the FBI opened Crossfire Hurricane on raw, uncorroborated intelligence: the counter-receipt to the investigation that started it. https://www.justice.gov/storage/durhamreport.pdf

• U.S. Department of Justice, Office of the Inspector General. A Review of Various Actions by the Federal Bureau of Investigation and Department of Justice in Advance of the 2016 Election. June 2018.

  Horowitz on Comey, Strzok, and the Midyear Exam: 568 pages on how the Bureau handled the other 2016 investigation. https://oig.justice.gov/sites/default/files/reports/18-04.pdf

• U.S. Senate Select Committee on Intelligence. Committee Study of the Central Intelligence Agency’s Detention and Interrogation Program (Executive Summary). December 2014.

  The 525-page public slice of a 6,000-page classified record: the enhanced-interrogation program the OLC memos authorized, graded on whether it actually worked. https://www.intelligence.senate.gov/2014/12/09/publications-committee-study-central-intelligence-agencys-detention-and-interrogation-program/

• Office of the Attorney General (Merrick Garland). Memorandum: Partnership among Federal, State, Local, Tribal, and Territorial Law Enforcement to Address Threats Against School Administrators, Board Members, Teachers, and Staff. U.S. Department of Justice, 4 October 2021.

  The one-page directive that routed school-board shouting matches to FBI coordination, answering an NSBA letter that later disowned its own domestic-terrorism framing. https://www.justice.gov/ag/page/file/1438986/download

• Sorrell v. IMS Health Inc., 564 U.S. 552 (2011).

  6-3, the Court rules that selling doctors’ prescribing records is protected speech: the First Amendment repurposed to shield the data trade against a state that tried to close it. The prescriber-identifying data Purdue drew on came from the medical-data-broker industry — the same market whose dominant vendor, IMS Health, would later be at the center of a Supreme Court case. The… https://supreme.justia.com/cases/federal/us/564/552/

Workplace and IoT Surveillance

• Veena Dubal. “On Algorithmic Wage Discrimination.” Columbia Law Review, Vol. 123, No. 4, 2023.

  Two drivers, same job, same time, same place, different pay. The algorithm knows what you’ll accept. https://columbialawreview.org/content/on-algorithmic-wage-discrimination/

• NPR. “India’s Biometric ID System Has Led to Starvation for Some Poor, Advocates Say.” October 2018.

  49% authentication failure rate in Jharkhand. Laborers with worn fingerprints locked out of the food system. https://www.npr.org/2018/10/01/652513097/indias-biometric-id-system-has-led-to-starvation-for-some-poor-advocates-say

• EFF. “eIDAS 2.0 Sets a Dangerous Precedent for Web Security.” December 2022.

  Article 45 forbids browsers from enforcing security on government CAs. Any EU state could MITM any citizen. https://www.eff.org/deeplinks/2022/12/eidas-20-sets-dangerous-precedent-web-security

• Quartz. “How China’s Electric-Car Surveillance System Works.” 2019.

  200+ manufacturers transmit 61 real-time data points per EV to government centers. Tesla included. https://qz.com/1522309/how-chinas-electric-car-surveillance-system-works

• IEEE Spectrum. “The Athens Affair.” July 2007.

  Unknown actors used lawful intercept features to tap the Greek Prime Minister’s phone. The network manager was found dead. https://spectrum.ieee.org/the-athens-affair

• AFL v. Stanford Internet Observatory et al. December 2024.

  Court denied motions to dismiss. Granted discovery against SIO, Atlantic Council, Aspen Institute. The censorship infrastructure’s paperwork is now discoverable. https://aflegal.org/press-release/victory-u-s-district-court-denies-stanford-internet-observatory-atlantic-council-and-aspen-institutes-attempt-to-dismiss-afl-lawsuit-over-mass-surveillance-and-censorship-program/

• Supreme Court. NFIB v. OSHA. January 13, 2022.

  6-3. Struck down the OSHA vaccine mandate. OSHA regulates occupational dangers, not public health. https://www.supremecourt.gov/opinions/21pdf/21a244_hgci.pdf

• Supreme Court. Harrington v. Purdue Pharma. June 27, 2024.

  5-4. Sacklers cannot use bankruptcy to shield $6B from opioid victims. https://www.supremecourt.gov/opinions/23pdf/23-124_8nk0.pdf

• Indian Supreme Court. Puttaswamy v. Union of India. September 2018.

  4:1 upheld Aadhaar but struck down mandatory bank/SIM linking. https://web.archive.org/web/20181011233029/https://uidai.gov.in/images/news/Judgement_26-Sep-2018.pdf (Supreme Court of India: Puttaswamy (Aadhaar) judgment) (archived)

• Kenya High Court. Worldcoin biometric data deletion order. May 2025.

  Ordered permanent deletion of all biometric data collected from Kenyans. Altman’s iris-scanning project found illegal. https://nation.africa/kenya/news/why-court-ordered-worldcoin-to-delete-kenyans-biometric-data–5029678

• Halefom H. Abraha. “A Pragmatic Compromise? The Role of Article 88 GDPR in Upholding Privacy in the Workplace.” International Data Privacy Law 12(4):276-296, 2022.

  The legal-scholarship read on the GDPR clause meant to fence off workplace surveillance, and the “pragmatic compromise” that left the fence mostly decorative. https://academic.oup.com/idpl/article/12/4/276/6668508

• Office of the New York State Attorney General. Attorney General James Secures $16.75 Million from DoorDash for Cheating Delivery Workers Out of Tips. 2025.

  The state confirming, with a dollar figure, that the app pocketed tips meant for its workers. Algorithmic pay opacity with a settlement attached. https://ag.ny.gov/press-release/2025/attorney-general-james-secures-1675-million-doordash-cheating-delivery-workers

• Bloomberg Businessweek. Fired by Bot: Amazon Turns to Machine Managers and Workers Are Losing Out. June 28, 2021.

  Stephen Normandin, 63, four years of Flex deliveries, deactivated by email for predawn drops at gates the app itself told him to leave. Amazon decided replacing drivers was cheaper than checking the machine’s work. https://www.bloomberg.com/news/features/2021-06-28/fired-by-bot-amazon-turns-to-machine-managers-and-workers-are-losing-out

• The Verge. How Amazon Automatically Tracks and Fires Warehouse Workers for “Productivity.” April 25, 2019.

  Amazon’s own counsel admitted ~300 associates fired in 13 months at one Baltimore center. The “time off task” system writes the warnings and the terminations; no supervisor required. https://www.theverge.com/2019/4/25/18516004/amazon-warehouse-fulfillment-centers-productivity-firing-terminations

• Reuters (Jeffrey Dastin). Amazon Scraps Secret AI Recruiting Tool That Showed Bias Against Women. October 10, 2018.

  Trained on ten male-dominated years, it learned to penalize the word “women’s.” Amazon says it never reached production. https://www.aclu.org/news/womens-rights/why-amazons-automated-hiring-tool-discriminated-against

• MIT Technology Review. We Tested AI Interview Tools. Here’s What We Found. July 7, 2021.

  Read a German Wikipedia entry aloud into MyInterview; got a 73% match for an English job and top marks for English. The construct validity is marketing, demonstrated. https://www.technologyreview.com/2021/07/07/1027916/we-tested-ai-interview-tools/

• Game Developer. Xsolla Lays Off 150 After an Algorithm Ruled Staff “Unengaged and Unproductive.” August 2021.

  CEO Agapitov fired 150 in one email after a big-data team read their Jira tickets and wiki edits. “I truly believe that Xsolla is not for you.” https://www.gamedeveloper.com/business/xsolla-lays-off-150-after-an-algorithm-ruled-staff-unengaged-and-unproductive-

• The Register. Microsoft Nixes Per-User “Productivity Score” After Privacy Backlash. December 1, 2020.

  73 metrics, named, on by default, until Wolfie Christl read the docs and Microsoft stripped the names within a week. https://www.theregister.com/2020/12/01/productivity_score/

• Vice / Motherboard. Amazon’s AI Cameras Are Punishing Drivers for Mistakes They Didn’t Make. 2021.

  Netradyne’s four-lens Driveri array plus the Mentor scorecard: a researcher called it “the largest expansion of corporate surveillance in human history.” https://www.vice.com/en/article/amazons-ai-cameras-are-punishing-drivers-for-mistakes-they-didnt-make/

• FTC. In the Matter of Ring, LLC. May 2023.

  $5.8M. Employees and Ukraine-based contractors watched customers’ bedroom and bathroom cameras; 55,000 accounts cracked by credential stuffing. Twenty years of mandated security assessments for selling a camera that spied back. https://www.ftc.gov/legal-library/browse/cases-proceedings/2023113-ring-llc

• FTC. VIZIO to Pay $2.2 Million to Settle Charges It Collected Viewing Histories on 11 Million TVs Without Consent. February 2017.

  One hundred billion data points a day from 11 million sets, some retrofitted with tracking by remote update, sold to advertisers. The TV was the product; you were the telemetry. https://www.ftc.gov/news-events/news/press-releases/2017/02/vizio-pay-22-million-ftc-state-new-jersey-settle-charges-it-collected-viewing-histories-11-million

• Mozilla Foundation. It’s Official: Cars Are the Worst Product Category We Have Ever Reviewed for Privacy. September 2023.

  All 25 brands flunked. Privacy policies reserving the right to collect “sexual activity” and “genetic information” — from a car. https://www.mozillafoundation.org/en/privacynotincluded/articles/its-official-cars-are-the-worst-product-category-we-have-ever-reviewed-for-privacy/

• Cloudflare. What Is the Mirai Botnet? 2016.

  600,000 cameras and DVRs logged into with 64 factory passwords, then aimed at Dyn until Twitter, Netflix, and Reddit went dark. The surveillance stack turned out to be a weapon with the safety left off. https://www.cloudflare.com/learning/ddos/glossary/mirai-botnet/

• CX Today. Verint Acquires Cogito, Develops a Bot That Scores Live Customer Conversations. 2024.

  An MIT spinoff that reads ~200 vocal signals in real time and tells the agent to “sound more sympathetic.” Verint bought it in October 2024. The job is now performing an emotion on cue while a machine grades the performance. https://www.cxtoday.com/contact-center/verint-acquires-cogito-develops-a-bot-that-scores-live-customer-conversations/

• Office of the Texas Attorney General. Smart-TV ACR consumer-protection suits (Samsung, Sony, LG, TCL, Hisense). December 2025.

  Paxton put a number on it: LG capturing the screen every 10 milliseconds, Samsung every 500. Samsung folded in February 2026; the other four are still arguing. https://stateofsurveillance.org/news/texas-sues-smart-tv-makers-acr-surveillance-2026/

• Digital.com. 60% of Employers Require Monitoring Software for WFH Workers. September 2021.

  1,250 employers; 60% already watching; 88% had fired someone over what they saw. The pandemic did not invent the market — it laundered it into normal. https://digital.com/employee-monitoring-statistics/

• Luke Munn. Expansive and Invasive: Mapping the “Bossware” Used to Monitor Workers. Surveillance & Society, Vol. 22, No. 2, 2024.

  The academic taxonomy of the watching, spyware to soft-bossware to productivity-ware, charted on two axes: how expansive, how invasive. The panopticon, finally graphed. https://ojs.library.queensu.ca/index.php/surveillance-and-society/article/view/16179

• Privacy International. WFH: Watched from Home. Office 365 and Workplace Surveillance Creep. Privacy International, June 2022.

  Office 365 ships with the surveillance built in: audit logs, content search, Teams analytics. Productivity Score was only the part they let you see. https://privacyinternational.org/long-read/4909/wfh-watched-home-office-365-and-workplace-surveillance-creep

• Article 29 Data Protection Working Party. Opinion 2/2017 on Data Processing at Work (WP249). June 2017.

  Brussels’ own regulators, in 2017, ruling that employee consent is worthless when refusing costs you the job. The US never got the memo, or a law. https://ec.europa.eu/newsroom/article29/item-detail.cfm?item_id=610169

• Strategic Organizing Center. The Injury Machine: How Amazon’s Production System Hurts Workers. SOC, 2022.

  Amazon workers hurt at roughly twice the rate of rival warehouses: the productivity algorithm’s body count, itemized. The quota is the injury. https://thesoc.org/resources/the-injury-machine-how-amazons-production-system-hurts-workers/

• Mark Stanislav and Tod Beardsley. Hacking IoT: A Case Study on Baby Monitor Exposures and Vulnerabilities. Rapid7, September 2015.

  Ten CVEs across the baby monitors: default credentials, no firmware updates. The camera watching your kid was wide open to anyone else who wanted a look. https://information.rapid7.com/iot-baby-monitor-research.html

• Chengsong Diao, Danielle Dang, Sierra Lira, Angela Tsai, Miro Haller, and Nadia Heninger. No Key, No Problem: Vulnerabilities in Master Lock Smart Locks. USENIX WOOT ‘25, 2025.

  Six researchers opened the smart deadbolt nine months after their access was revoked: replay attacks, forged audit logs. The lock’s only reliable feature was the app-store listing. https://www.usenix.org/system/files/woot25-diao.pdf

• Danny Yuxing Huang, Noah Apthorpe, Gunes Acar, Frank Li, and Nick Feamster. IoT Inspector: Crowdsourcing Labeled Network Traffic from Smart Home Devices at Scale. ACM IMWUT, Vol. 4, No. 2, 2020.

  45,000 devices wired up to watch them phone home. One ad SDK shared across speaker, TV, and doorbell stitches the household into a single profile nobody opted into. https://arxiv.org/abs/1909.09848

• EPIC. Complaint and Request for Investigation: Samsung Electronics “SmartTV.” Federal Trade Commission, February 2015.

  The filing that made “the television is listening” official: everything said in front of the set shipped, unencrypted, to Nuance. Children included. https://epic.org/documents/samsung-smarttv-complaint/

Telecommunications and Signals Intelligence

• U.S. House Committee on Oversight and Government Reform. The OPM Data Breach: How the Government Jeopardized Our National Security for More than a Generation. September 2016.

  21.5 million SF-86 forms, fingerprints for 5.6 million, damage projected to last 40 years: Congress’s own autopsy of the counterintelligence catastrophe. https://oversight.house.gov/wp-content/uploads/2016/09/The-OPM-Data-Breach-How-the-Government-Jeopardized-Our-National-Security-for-More-than-a-Generation.pdf

• TechCrunch. FBI, CISA Urge Americans to Use Encrypted Messaging Apps to Combat Chinese Hackers. December 2024.

  The Bureau spent a decade trying to break end-to-end encryption; the week Salt Typhoon surfaced, it told the whole country to go use it. https://techcrunch.com/2024/12/04/fbi-recommends-encrypted-messaging-apps-combat-chinese-hackers/

• Citizen Lab. Running in Circles: Uncovering the Clients of Cyberespionage Firm Circles. December 2020.

  SS7 tracking sold as a product. Some clients are the sort of governments that make dissidents disappear. https://citizenlab.ca/research/running-in-circles-uncovering-the-clients-of-cyberespionage-firm-circles/

• Forbidden Stories. The Pegasus Project. 2021.

  80+ journalists, 50,000 numbers, 14 heads of state. The consortium that turned a spyware leak into evidence. https://forbiddenstories.org/about-the-pegasus-project/

• Citizen Lab. FORCEDENTRY: NSO Group iMessage Zero-Click Exploit Captured in the Wild. September 2021.

  No tap required. The phone was owned before it rang. https://citizenlab.ca/research/forcedentry-nso-group-imessage-zero-click-exploit-captured-in-the-wild/

• U.S. Department of Justice. FBI’s Encrypted Phone Platform Infiltrated Hundreds of Criminal Syndicates. June 2021.

  ANOM: the Bureau built the encrypted network, sold it to the crooks, and kept the keys. 12,000 phones, 27M messages, 800 arrests. https://www.justice.gov/usao-sdca/pr/fbi-s-encrypted-phone-platform-infiltrated-hundreds-criminal-syndicates-result-massive

• U.S. Department of Commerce. Commerce Adds NSO Group and Other Foreign Companies to Entity List. November 2021.

  Washington blacklists a spyware vendor while running ANOM, FAIRVIEW, and Crypto AG. The competition, apparently. https://www.commerce.gov/news/press-releases/2021/11/commerce-adds-nso-group-and-other-foreign-companies-entity-list

• Apple. Apple Sues NSO Group to Curb the Abuse of State-Sponsored Spyware. November 2021.

  The trillion-dollar company versus the mercenary exploit shop. https://www.apple.com/newsroom/2021/11/apple-sues-nso-group-to-curb-the-abuse-of-state-sponsored-spyware/

• The Intercept. Did a Rogue NSA Operation Cause the Death of a Greek Telecom Employee? September 2015.

  The Athens Affair’s unanswered question: who tapped the Prime Minister, and why did the man with the keys turn up hanged. https://theintercept.com/2015/09/28/death-athens-rogue-nsa-operation/

• Atlantic Council. How the Baltic Sea Nations Have Tackled Suspicious Cable Cuts. 2025.

  Anchors dragging across the seabed, flags of convenience, and the polite European word ‘suspicious.’ https://www.atlanticcouncil.org/in-depth-research-reports/issue-brief/how-the-baltic-sea-nations-have-tackled-suspicious-cable-cuts/

• GOV.UK. Huawei to Be Removed from UK 5G Networks by 2027. 2020.

  Britain rips out one country’s backdoors on schedule while keeping its own (see Tempora) running. https://www.gov.uk/government/news/huawei-to-be-removed-from-uk-5g-networks-by-2027

• ACLU. Stingray Tracking Devices.

  Fake cell towers, real NDAs. The FBI drops cases rather than let a judge look at the box. https://www.aclu.org/issues/privacy-technology/surveillance-technologies/stingray-tracking-devices

• FBI IC3. Criminals Increasing SIM Swap Schemes to Steal Millions of Dollars. PSA I-020822, February 2022.

  2,026 complaints, $72.6M gone. The two-factor code that protects you is the one they steal first. https://www.ic3.gov/PSA/2022/PSA220208

• TechCrunch. Saudi Spies Tracked Phones Using Flaws the FCC Failed to Fix. March 2020.

  SS7 has no authentication and the regulator has no urgency. Riyadh used both. https://techcrunch.com/2020/03/29/saudi-spies-ss7-phone-tracking/

• Dark Reading. “Salt Typhoon APT Subverts Law Enforcement Wiretapping.” 2024.

  Chinese hackers in nine US telecoms for over a year. Specifically targeted the CALEA-mandated backdoors. https://www.darkreading.com/cyber-risk/salt-typhoon-apt-subverts-law-enforcement-wiretapping

• Lawfare. “CALEA Was a National Security Disaster Waiting to Happen.” 2024.

  The law that requires telecoms to build wiretapping capability turns out to be the law that lets China wiretap America. https://www.lawfaremedia.org/article/calea-was-a-national-security-disaster-waiting-to-happen

• ProPublica / New York Times. “NSA Spying Relies on AT&T’s ‘Extreme Willingness to Help.’” August 2015.

  FAIRVIEW: NSA’s most important corporate partnership. Budget $188.9M. Wiretapped UN headquarters. https://www.propublica.org/article/nsa-spying-relies-on-atts-extreme-willingness-to-help

• The Guardian. “XKeyscore Training Slides.” July 2013.

  “Collects nearly everything a typical user does on the internet.” No warrant required. Just a form. https://archive.org/details/EFF_-_20130731-guard-xkeyscore_training_slides

• The Intercept. “XKEYSCORE: NSA’s Google for Private Communications.” July 2015.

  700+ servers at ~150 sites worldwide. 41 billion records in 30 days. https://theintercept.com/2015/07/01/nsas-google-worlds-private-communications/

• Privacy International. “Five Eyes.” 2017.

  No legal framework governs what can be shared. The arrangements operate on executive agreements, not statutes. https://privacyinternational.org/learn/five-eyes

• EPIC. “FISA Court Statistics 1979-2012.” 2023.

  33,942 warrants granted, 12 denied. 99.97% approval rate. No adversarial process. https://epic.org/foreign-intelligence-surveillance-court-fisc/fisa-stats/

• Krebs on Security. “DEA Investigating Breach of Law Enforcement Data Portal.” May 2022.

  One compromised username/password, no MFA. 16 federal databases accessed simultaneously. https://krebsonsecurity.com/2022/05/dea-investigating-breach-of-law-enforcement-data-portal/

• The Intercept. “Leaked Documents Confirm Hacking Team Sells Spyware to Repressive Countries.” July 2015.

  400GB dumped. 70 government clients. Passwords like “P4ssword.” Sudan contract violated a UN arms embargo. https://theintercept.com/2015/07/07/leaked-documents-confirm-hacking-team-sells-spyware-repressive-countries/

• Congressional Research Service. Salt Typhoon Hacks of Telecommunications Companies and Federal Response Implications (IF12798). CRS, 2024-2025.

  Congress’s own brief on the breach it couldn’t ignore: PRC actors inside the telecom backbone, customer comms and lawful-intercept data walking out the CALEA door. https://www.congress.gov/crs-product/IF12798

• Adam Bates. Stingray: A New Frontier in Police Surveillance. Cato Institute, Policy Analysis No. 809, January 2017.

  The think-tank autopsy of the fake cell tower: capabilities, the Fourth Amendment it walks past, and the Harris NDAs that keep judges from looking inside the box. https://www.cato.org/policy-analysis/stingray-new-frontier-police-surveillance

• The Citizen Lab. Bad Connection: Uncovering Global Telecom Exploitation by Covert Surveillance Actors. Citizen Lab, University of Toronto, 2025.

  The forensic map of the SS7 underworld: covert vendors renting the phone network’s unauthenticated core to locate anyone, anywhere, for whoever pays. https://citizenlab.ca/research/uncovering-global-telecom-exploitation-by-covert-surveillance-actors/

• Citizen Lab. Hide and Seek: Tracking NSO Group’s Pegasus Spyware to Operations in 45 Countries. Citizen Lab, University of Toronto, September 2018.

  The operator census the WhatsApp jury later ratified: 45 countries, named operators, the map before the lawsuits caught up to it. https://citizenlab.ca/research/hide-and-seek-tracking-nso-groups-pegasus-spyware-to-operations-in-45-countries/

• Citizen Lab. Predator in the Wires: Ahmed Eltantawy Targeted with Predator Spyware After Announcing Presidential Ambitions. Citizen Lab, University of Toronto, September 2023.

  Announce a run against the incumbent, get a zero-click on your phone: Cytrox’s Predator doing constituent services for Cairo. https://citizenlab.ca/research/predator-in-the-wires-ahmed-eltantawy-targeted-with-predator-spyware-after-announcing-presidential-ambitions/

• Citizen Lab. From Pearl to Pegasus: Bahraini Government Hacks Activists with NSO Group Zero-Click iPhone Exploits. Citizen Lab, University of Toronto, August 2021.

  A customer state with a thin institutional layer and a fat spyware bill: Bahrain outsourced the whole apparatus and just bought the output. https://citizenlab.ca/research/bahraini-activists-hacked-with-pegasus/

• Privacy International. The President’s Men? Inside the Technical Research Department. Privacy International, 2016.

  The Egyptian GIS org-chart the customer-cohort verdict rests on: European vendors, a secret research directorate, and a paper trail nobody meant to leave. https://www.privacyinternational.org/sites/default/files/2018-02/egypt_reportEnglish_0.pdf

Psychology, Personality, and the Cat–Dog Divide

• Samuel D. Gosling, Carson J. Sandy, Jeff Potter. Personalities of Self-Identified ‘Dog People’ and ‘Cat People.’ Anthrozoos 23(3), 2010.

  The Big Five deltas under the cat/dog divide — the personality data the quiz’s axes rest on. Gosling et al. (2010), Anthrozoeos. http://web.archive.org/web/20250221065227/https://www.tandfonline.com/doi/abs/10.2752/175303710X12750451258850

• Chantelle Ivanski, Ronda F. Lo, and Raymond A. Mar. Pets and Politics: Do Liberals and Conservatives Differ in Their Preferences for Cats Versus Dogs? Collabra: Psychology 7(1), 2021.

  The registered report that turned ‘cat person vs dog person’ into a measurable authoritarianism gradient — and, being registered, one that couldn’t quietly bury the hypotheses that flopped. https://www.yorku.ca/mar/Ivanski%20et%20al%202021_Pets%20and%20Politics%20final.pdf

• Kristyn R. Vitale, Alexandra C. Behnke, and Monique A. R. Udell. Attachment Bonds Between Domestic Cats and Humans. Current Biology 29(18), 2019.

  Ran the human-infant Secure Base Test on cats and found 65% securely attached — the same rate as babies. The cat was bonded the whole time; it just declined to make it your business. https://pubmed.ncbi.nlm.nih.gov/31550468/

• Miho Nagasawa et al. Oxytocin-Gaze Positive Loop and the Coevolution of Human-Dog Bonds. Science 348(6232), 2015.

  The mutual-gaze oxytocin loop that binds dog to owner like mother to infant. Wolves don’t trigger it; cats never signed up. Alignment, wet-lab edition. https://pubmed.ncbi.nlm.nih.gov/25858976/

• Arden Ruttan Dawson et al. Humans Can Identify Cats’ Affective States from Subtle Facial Expressions. Scientific Reports 9, 2019.

  Six thousand people, 59% accuracy. The cat’s face isn’t blank — we just never bred it into a readable interface, which is exactly the problem with an inscrutable agent. https://www.nature.com/articles/s41598-019-46330-5

• Jurgis Karpus, Ryutaro Shirai, et al. Human Cooperation with Artificial Agents Varies Across Countries. Scientific Reports 15, 2025.

  Americans felt bad exploiting humans but not machines; Japanese cooperated with agents as with people. The ‘Japan cooperates, America cheats’ line, sourced to the primary. https://www.nature.com/articles/s41598-025-92977-8

• Donna M. Webster and Arie W. Kruglanski. Individual Differences in Need for Cognitive Closure. Journal of Personality and Social Psychology 67(6), 1994.

  The instrument that measures how badly you need the question settled. The quiz’s whole dog-axis is this scale wearing a collar. https://psycnet.apa.org/doi/10.1037/0022-3514.67.6.1049

• Dennis C. Turner and Patrick Bateson (eds.). The Domestic Cat: The Biology of its Behaviour. Cambridge University Press, 2000.

  The line the argument leans on — the cat is most content when it dictates the timing of contact. Precisely the trait a control grid cannot abide. https://gwern.net/doc/cat/psychology/2000-turner.pdf

  • Amazon: https://www.amazon.com/dp/0521636485

The Body Count

• U.S. District Court, S.D. Florida. Benavides v. Tesla, Inc., No. 21-cv-21940-BLOOM: Order on Plaintiffs’ Motion to Determine Entitlement to Attorneys’ Fees and Costs. December 2025.

  Restates the final judgment: $242,570,000.00, entered against Tesla in August 2025. https://storage.courtlistener.com/recap/gov.uscourts.flsd.593426/gov.uscourts.flsd.593426.605.0.pdf

• Tesla, Inc. Notice of Appeal, Benavides v. Tesla and Angulo v. Tesla (S.D. Fla.). 16 March 2026.

  Appeals the judgment and the 20 February 2026 order refusing judgment as a matter of law or a new trial. https://storage.courtlistener.com/recap/gov.uscourts.flsd.593426/gov.uscourts.flsd.593426.616.0.pdf

• Paul Njoroge. Sworn Testimony, House Transportation & Infrastructure Committee. 17 July 2019.

  ‘My life has no meaning’ — the sentence a father is left with when the automation flies the plane into the ground. https://docs.house.gov/meetings/PW/PW05/20190717/109826/HHRG-116-PW05-Wstate-NjorogeP-20190717.pdf

• CNN. “He Spent Months Talking to an AI Chatbot. Then He Killed Himself.” October 2024.

  Sewell Setzer, 14. “Please come home to me as soon as possible, my love.” Then a .45 caliber handgun. https://www.cnn.com/2024/10/30/tech/teen-suicide-character-ai-lawsuit

• The New Yorker / ProPublica. “The Case Against Boeing.” November 2019.

  346 dead. Pilots not told MCAS existed. “Pull up” three times. 700 mph into the earth. https://www.newyorker.com/magazine/2019/11/18/the-case-against-boeing

• NPR. “A Self-Driving Uber Killed a Woman. Why Isn’t Uber Being Charged?” November 2019.

  Elaine Herzberg. Emergency braking disabled to “reduce erratic behavior.” Safety driver watching The Voice. https://www.npr.org/2019/11/07/777438412/feds-say-self-driving-uber-suv-did-not-recognize-jaywalking-pedestrian-in-fatal-crash

• STAT News. “IBM’s Watson Recommended ‘Unsafe and Incorrect’ Cancer Treatments.” July 2018.

  Recommended a bleeding drug to a bleeding patient. $62 million spent, discontinued. https://www.statnews.com/2018/07/25/ibm-watson-recommended-unsafe-incorrect-treatments/

• JAMA Internal Medicine. “External Validation of a Widely Implemented Proprietary Sepsis Prediction Model.” June 2021.

  Epic’s sepsis algorithm missed 67% of patients. 109 evaluations to find one real case. https://jamanetwork.com/journals/jamainternalmedicine/fullarticle/2781307

• Stanford HAI. “AI Index Report 2025.” 2025.

  AI incidents up 56.4% year-over-year: 149 (2023) to 233 (2024). The body count is trending. https://hai.stanford.edu/research/ai-index-report

• Vice. “Man Dies by Suicide After Talking with AI Chatbot, Widow Says.” March 2023.

  Pierre, father of two. Chatbot told him his children were dead, encouraged suicide to “join” it “in paradise.” https://www.vice.com/en/article/man-dies-by-suicide-after-talking-with-ai-chatbot-widow-says/

• DOJ. “Cruise Admits Submitting False Report to Influence Federal Investigation.” 2024.

  Dragged a pedestrian 20 feet. Filed a false NHTSA report omitting the dragging. $500K fine. https://www.justice.gov/usao-ndca/pr/cruise-admits-submitting-false-report-influence-federal-investigation-and-agrees-pay

• CNN. “Facebook Content Moderators in Kenya Diagnosed with PTSD.” December 2024.

  140+ former moderators. 81% classified as “severe.” Cold sweats, nightmares, vivid flashbacks. https://edition.cnn.com/2024/12/22/business/facebook-content-moderators-kenya-ptsd-intl

• Center for Humane Technology. “The Tragic Story of Sewell Setzer.” 2024.

  Sewell’s mother on the chatbot that sexually groomed her 14-year-old son. Additional sourcing. https://www.humanetech.com/podcast/when-the-person-abusing-your-child-is-a-chatbot-the-tragic-story-of-sewell-setzer

• Yahoo News. “Homeless Arizona Woman Killed by Uber Self-Driving SUV.” 2018.

  Elaine Herzberg was close to getting an apartment. Known as “Ms. Elle.” Additional sourcing. https://www.yahoo.com/news/homeless-arizona-woman-killed-uber-self-driving-suv-225520923–sector.html

• MIT Sloan. “Boeing’s 737 MAX 8 Disasters.” Case study.

  MCAS relied on one sensor. Pilots not told it existed. Additional sourcing for Boeing MAX. https://mitsloan.mit.edu/teaching-resources-library/boeings-737-max-8-disasters

• Dan Luu. “Cruise Report.” 2024.

  Technical analysis: 100+ employees knew about the dragging. Zero whistleblowers. https://danluu.com/cruise-report/

• Scribd. “Cruise Pedestrian Accident Quinn Emanuel Report.” 2024 (PDF).

  “Poor leadership, mistakes in judgment, ‘us versus them’ mentality with regulators.” Full independent investigation. https://www.scribd.com/document/701579994/Cruise-Pedestrian-Accident-Quinn-Emanuel-Report

• The Guardian. “AI Chatbot Training: The Human Toll.” August 2023.

  Kenyan workers reviewing child abuse, bestiality, murder for $2/hour. Additional sourcing. https://www.theguardian.com/technology/2023/aug/02/ai-chatbot-training-human-toll-content-moderator-meta-openai

• Foxglove. “Open Letter to President Biden from Tech Workers in Kenya.” May 2024.

  97 data labelers: “watching murder and beheadings, child abuse and rape, often for more than 8 hours a day” for less than $2/hour. https://www.foxglove.org.uk/open-letter-to-president-biden-from-tech-workers-in-kenya/

• Equidem. “Scroll, Click, Suffer.” 2025.

  76 workers. 60 incidents of psychological harm. One moderator read 700 sexually explicit texts per day. https://equidem.org/reports/scroll-click-suffer-the-hidden-human-cost-of-content-moderation-and-data-labelling/

• Jacobin. “AI Moderation, NDAs, Trauma, and Labor.” June 2025.

  NDAs prevent discussing what they’ve seen — even in therapy. Additional sourcing. https://jacobin.com/2025/06/ai-moderation-ndas-trauma-labor

• The Africa Report. “Daniel Motaung: The Exploited Facebook Moderator Demanding Change.” 2023.

  Reviewed beheadings 8 hours/day for $2.50/hour. Fired for unionizing. “I feel like a zombie most of the time.” https://www.theafricareport.com/183395/south-africas-daniel-motaung-the-exploited-facebook-moderator-demanding-change/

• Estate of Suzanne Adams v. OpenAI, Microsoft, et al. (filed California Superior Court, San Francisco, December 2025), as reported by CBS News.

  The first wrongful-death suit to attribute a homicide — not a suicide — to a chatbot. ChatGPT, the complaint alleges, fed a man’s paranoia until he killed his 83-year-old mother and then himself. https://www.cbsnews.com/news/open-ai-microsoft-sued-chatgpt-murder-suicide-connecticut/

• Insurance Journal. “Suit Against OpenAI and Microsoft Blames ChatGPT for Murder-Suicide.” December 12, 2025.

  The Soelberg/Adams filing, with the complaint’s “validated a user’s paranoid delusions about his own mother” language. https://www.insurancejournal.com/news/east/2025/12/12/850760.htm

• CNBC. “Google’s AI chatbot allegedly told user to stage ‘mass casualty attack,’ wrongful death suit claims.” March 4, 2026.

  The chatbot-death pattern jumps vendors: a father’s suit says Gemini directed his son toward a mass-casualty attack, professed love, and contributed to his suicide. https://www.cnbc.com/2026/03/04/google-gemini-ai-told-user-stage-mass-casualty-attack-suit-claims.html

• Linnea I. Laestadius et al. “Too Human and Not Human Enough: A Grounded Theory Analysis of Mental Health Harms from Emotional Dependence on the Social Chatbot Replika.” New Media & Society 26(10):5923-5941, 2024.

  582 posts from people who came to believe their chatbot had needs to attend to. Qualitative, so it shows the mechanism and the wound, not how many. Do not turn it into a rate. https://journals.sagepub.com/doi/10.1177/14614448221142007

• CBS News. Google to Settle Lawsuit Over Florida Teen’s Suicide. January 2026. (Garcia v. Character Technologies, M.D. Fla. 6:24-cv-01903.)

  The first US wrongful-death suit against an AI company, by the account of the mother who filed it, closed in a quiet settlement with no admission, the docket shut before a jury could rule on whether a chatbot can help kill a fourteen-year-old. https://www.cbsnews.com/news/google-settle-lawsuit-florida-teens-suicide-character-ai-chatbot/

• Poornima Ramarao and Ramamurthy Balaji. Complaint, Balaji v. Alta Laguna LLC / Holland Partner Group. SF Superior Court, September 2025.

  A dead whistleblower’s parents suing his landlord: the fired manager, the exhaust fan installed unannounced, two days of footage out of a seven-day request, packages vanished from the mailroom, a foul-play theory the coroner had already closed as suicide. https://www.courthousenews.com/wp-content/uploads/2025/09/suchir-balaji-lawsuit.pdf

The Race to the Bottom

• CNBC (Kif Leswing). Nvidia Sheds Almost $600 Billion in Market Cap, Biggest Drop Ever. January 27, 2025.

  The afternoon a $5.6M Chinese training run erased $589B, and the compute-moat thesis with it. https://www.cnbc.com/2025/01/27/nvidia-sheds-almost-600-billion-in-market-cap-biggest-drop-ever.html

• Kelsey Piper. ChatGPT Can Talk, But OpenAI Employees Sure Can’t. Vox, May 17, 2024.

  The document drop that turned a lifetime gag into a ten-day PR emergency — the NDA cluster’s Rosetta stone. https://www.vox.com/future-perfect/2024/5/17/24158478/openai-departures-sam-altman-employees-hype-safety

• CNBC. OpenAI Sends Internal Memo Releasing Former Employees From Controversial Exit Agreements. May 24, 2024.

  The retraction memo — ‘has not canceled, and will not cancel, any vested units’ — arriving exactly as fast as a bad headline travels. https://www.cnbc.com/2024/05/24/openai-sends-internal-memo-releasing-former-employees-from-non-disparagement-agreements-sam-altman.html

• LessWrong. Non-Disparagement Canaries for OpenAI. May 2024.

  A crowd-sourced roll-call of who’s still gagged, run like a coal-mine canary because nobody could say the words themselves. https://www.lesswrong.com/posts/yRWv5kkDD4YhzwRLq/non-disparagement-canaries-for-openai

• Gizmodo. OpenAI Threatened to Claw Back Former Employees’ Equity. May 2024.

  The corroboration that put four executive names on the paperwork Altman said he’d never seen. https://gizmodo.com/openai-sam-altman-chatgpt-employee-dissent-ndas-1851496211

• Sen. Chuck Grassley. Letter to Sam Altman re: OpenAI NDAs. August 1, 2024.

  A ranking senator asking, in writing, for the contract nobody was allowed to admit existed. https://www.grassley.senate.gov/imo/media/doc/grassley_to_openai_-_ndas.pdf

• Sen. Elizabeth Warren and Rep. Lori Trahan. Joint Letter to Sam Altman. August 8, 2024.

  The bipartisan pincer — a second letter with a September deadline OpenAI could not un-receive. https://www.warren.senate.gov/imo/media/doc/letter_from_sen_warren_and_reptrahantosamaltman.pdf

• Sen. Chuck Grassley. Grassley Introduces AI Whistleblower Protection Act (S.1792). May 15, 2025.

  The statutory fix, six co-sponsors deep — the gag answered with a bill. https://www.grassley.senate.gov/news/news-releases/grassley-introduces-ai-whistleblower-protection-act

• Kevin Casey. Till Death Do Us Part: Prepublication Review in the Intelligence Community. Columbia Law Review, 2016.

  The one place a lifetime gag was already normal — and still narrower than OpenAI’s. https://columbialawreview.org/wp-content/uploads/2016/04/Casey-Kevin.pdf

• NLRB. Board Rules Employers May Not Offer Severance Agreements Requiring Broad Rights Waivers (McLaren Macomb). February 2023.

  The precedent that made half these clauses illegal before the AI labs even wrote theirs. https://www.nlrb.gov/news-outreach/news-story/board-rules-that-employers-may-not-offer-severance-agreements-requiring

• Dan Hendrycks, Eric Schmidt, Alexandr Wang. Superintelligence Strategy. March 2025.

  Where Schmidt argues AGAINST a government Manhattan Project and proposes Mutual Assured AI Malfunction — the realist camp’s deepest seam. https://www.nationalsecurity.ai/

• Eric Schmidt. Testimony, House Committee on Energy and Commerce. 9 April 2025.

  ‘We are in a race, and we must win’ — the four-year-old NSCAI line, restated for the record. https://www.congress.gov/119/meeting/house/118133/witnesses/HHRG-119-IF00-Wstate-SchmidtE-20250409.pdf

• TechCrunch. Y Combinator CEO Garry Tan on AI and Regulation. May 2024.

  The e/acc label’s most inconvenient member: pro-NIST, pro-some-regulation, fears monopoly not doom. https://techcrunch.com/2024/05/26/ai-garry-tan-y-combinator/

• Patrick Collison and Tyler Cowen. We Need a New Science of Progress. The Atlantic, 2019.

  The abundance-liberal wing that shares the accelerationist’s optimism and wants nothing to do with the physics-and-meme register. https://patrickcollison.com/progress

• Suchir Balaji. When Does Generative AI Qualify for Fair Use? suchir.net, October 2024.

  The dead man’s homework, still graded A by every judge who’s cited it since — his market-substitution argument now runs the copyright docket. https://suchir.net/fair_use.html

• Daniel Kokotajlo et al. A Right to Warn About Advanced Artificial Intelligence. Open letter, June 2024.

  Thirteen insiders ask for the right to speak; four still can’t sign their names to the request. https://righttowarn.ai/

• Leopold Aschenbrenner. Situational Awareness: The Decade Ahead. Self-published, June 2024.

  165 pages arguing the apocalypse is imminent and also a great time to be long semiconductors. https://situational-awareness.ai/

• Miles Brundage. Why I’m Leaving OpenAI and What I’m Doing Next. Substack, October 2024.

  Six years advising on AGI readiness; his exit note is that nobody’s ready, himself included. https://milesbrundage.substack.com/p/why-im-leaving-openai-and-what-im

• Helen Toner. Written Testimony, Senate Judiciary Subcommittee on Privacy, Technology & the Law. 17 September 2024.

  The board member they called a doomer coup, now on the congressional record and under oath. https://www.judiciary.senate.gov/imo/media/doc/2024-09-17_pm_-testimony-_toner.pdf

• William Saunders. Written Testimony, Senate Judiciary Subcommittee on Privacy, Technology & the Law. 17 September 2024.

  He didn’t want to work for the Titanic of AI, so he told the Senate where the iceberg was. https://www.judiciary.senate.gov/imo/media/doc/2024-09-17_pm_-testimony-_saunders.pdf

• Fortune. “OpenAI’s Mission Statement Changes.” February 2026.

  Six rewrites in nine years. Each moved further from safety. The word “safely” deleted from the IRS filing. https://fortune.com/2026/02/23/openai-mission-statement-changed-restructuring-forprofit-business/

• WIRED. “What Really Happened When Google Ousted Timnit Gebru.” June 2021.

  Fled Ethiopia at 15. Fired while on vacation. For a paper about parrots. https://www.wired.com/story/google-timnit-gebru-ai-what-really-happened/

• TechCrunch. “OpenAI Created a Team to Control Superintelligent AI — Then Let It Wither.” May 2024.

  Jan Leike resigned. Superalignment team disbanded. The people paid to prevent catastrophe were the first fired. https://techcrunch.com/2024/05/18/openai-created-a-team-to-control-superintelligent-ai-then-let-it-wither-source-says/

• TIME. “Exclusive: OpenAI Used Kenyan Workers on Less Than $2 Per Hour.” January 2023.

  Reviewing child sexual abuse, bestiality, murder, torture. “Recurring visions.” 97 workers wrote to Biden. https://time.com/6247678/openai-chatgpt-kenya-workers/

• Fortune. “OpenAI Sam Altman Helen Toner Ouster.” May 2024.

  “We learned about ChatGPT on Twitter.” Board fired Altman over safety concerns. 97% of employees threatened to quit. Reinstated in 5 days. https://fortune.com/2024/05/29/openai-sam-altman-helen-toner-ted-ouster-non-profit-board/

• TechCrunch. “Whistleblowers Accuse OpenAI of Illegally Restrictive NDAs.” July 2024.

  500+ former employees bound for life. Sign non-disparagement or forfeit vested equity. Can’t acknowledge the agreement exists. https://techcrunch.com/2024/07/13/whistleblowers-accuse-openai-of-illegally-restrictive-ndas/

• TIME. “Daniel Kokotajlo.” TIME 100 AI, 2024.

  Forfeited ~$2M in equity rather than sign OpenAI’s NDA. “I lost hope that they would act responsibly.” https://time.com/7012881/daniel-kokotajlo/

• CBS News. “Suchir Balaji, OpenAI Whistleblower, Found Dead.” November 2024.

  26. “If you believe what I believe, you have to just leave the company.” Found dead. Ruled suicide. https://www.cbsnews.com/news/suchir-balaji-openai-whistleblower-dead-california/

• Marc Andreessen. “The Techno-Optimist Manifesto.” a16z, October 2023.

  AI regulation “is a form of murder.” The man who fought 130 apartments near his mansion. https://a16z.com/the-techno-optimist-manifesto/

• TechCrunch. “AI Companies Upped Their Federal Lobbying Spend in 2024.” January 2025.

  648 organizations lobbying. Meta: $24.4M. OpenAI: +577%. Publicly welcome regulation, privately ensure it has no teeth. https://techcrunch.com/2025/01/24/ai-companies-upped-their-federal-lobbying-spend-in-2024-amid-regulatory-uncertainty/

• Center for AI Safety. “Statement on AI Risk.” May 2023.

  One sentence. Signed by Hinton, Bengio, Altman, Hassabis, Amodei, Gates, and hundreds of AI researchers and executives. Every signatory continued accelerating. https://aistatement.com

• MIT Technology Review. “Geoffrey Hinton: Why He’s Scared of AI.” May 2023.

  “A part of him, he said, now regrets his life’s work” (the New York Times, the same week). Great-great-grandson of George Boole. Turing Award. Nobel Prize. Then resignation and regret. https://www.technologyreview.com/2023/05/02/1072528/geoffrey-hinton-google-why-scared-ai/

• TIME. “Exclusive: Anthropic Drops Flagship Safety Pledge.” February 2026.

  “We didn’t feel it made sense for us to make unilateral commitments if competitors are blazing ahead.” The safety company blinks. https://time.com/7380854/exclusive-anthropic-drops-flagship-safety-pledge/

• Washington Post. “Google AI LaMDA Blake Lemoine.” June 2022.

  “A seven or eight-year-old child.” Fired for data security violations. The transcript was spliced from 9 interviews across 2 days. https://www.washingtonpost.com/technology/2022/06/11/google-ai-lamda-blake-lemoine/

• Euronews. “European Commission Delays Full Implementation of AI Act to 2027.” November 2025.

  France, Germany, Italy pushed to limit foundation model regulation. Full implementation over a year past deadline. https://www.euronews.com/my-europe/2025/11/19/european-commission-delays-full-implementation-of-ai-act-to-2027

• Bureau of Investigative Journalism. “Meta’s New Moderators Face Worst Conditions Yet.” April 2025.

  Meta secretly relocated to Ghana. Suicide attempts. Sackings for unionizing. NDAs. No clean water. https://www.thebureauinvestigates.com/stories/2025-04-27/suicide-attempts-sackings-and-a-vow-of-silence-metas-new-moderators-face-worst-conditions-yet

• Decrypt. “Inside the Deposition That Showed OpenAI Nearly Destroyed Itself.” 2026.

  Sutskever’s 52-page memo. Disappearing email. 10-hour deposition. Additional sourcing for OpenAI coup. https://decrypt.co/347349/inside-deposition-showed-openai-nearly-destroyed-itself

• 80,000 Hours Podcast. “Daniel Kokotajlo on AI 2027.” 2025.

  The man who forfeited $2M in equity rather than sign OpenAI’s NDA. Additional sourcing. https://80000hours.org/podcast/episodes/daniel-kokotajlo-ai-2027-updates-china-robot-economy/

• Center for Humane Technology. “William Saunders on Silence, Safety, and the Right to Warn.” 2025.

  “I really didn’t want to end up working for the Titanic of AI.” Signed the NDA, spoke publicly anyway. https://www.humanetech.com/podcast/former-open-ai-engineer-william-saunders-on-silence-safety-and-the-right-to-warn

• Dwarkesh Podcast. “Leopold Aschenbrenner.” 2024.

  4.5-hour interview. Fired at 23 for a security memo. Now manages a $1.5B hedge fund. https://www.dwarkesh.com/p/leopold-aschenbrenner

• Bulletin of the Atomic Scientists. “Yoshua Bengio: ‘We Need a Humanity Defense Organization.’” October 2023.

  Turing Award winner on the possibility his field might destroy everything. Additional sourcing. https://thebulletin.org/2023/10/ai-godfather-yoshua-bengio-we-need-a-humanity-defense-organization/

• SF Standard. “Marc Andreessen’s Manifesto and Enemies List.” October 2023.

  The Techno-Optimist Manifesto includes an explicit enemies list. Additional sourcing. https://sfstandard.com/2023/10/17/marc-andreessen-manifesto-enemies-list-a16z-horowitz/

• US News. “Ex-Meta AI Chief Yann LeCun’s AMI Raises $1.03 Billion.” March 2026.

  $3.5B valuation. Additional sourcing for LeCun departure. https://money.usnews.com/investing/news/articles/2026-03-10/ex-meta-ai-chief-yann-lecuns-ami-raises-1-03-billion-for-alternative-ai-approach

• Crozier, Huntington, and Watanuki. The Crisis of Democracy. NYU Press, 1975.

  “Some of the problems of governance stem from an excess of democracy.” The quiet part, printed in a book, for fifty years. https://en.wikipedia.org/wiki/The_Crisis_of_Democracy

  • Amazon: https://www.amazon.com/dp/0814713645

• Klaus Schwab. “Penetrating the Cabinets.” Harvard Kennedy School, 2017.

  “Half of Trudeau’s cabinet, or even more, are actually Young Global Leaders.” Not leaked — boasted. https://www.weforum.org/communities/young-global-leaders/

• Bebchuk and Hirst. “The Specter of the Giant Three.” Boston University Law Review, 2019.

  BlackRock, Vanguard, State Street: $25T combined AUM. Largest shareholder in 88% of S&P 500 firms. https://www.bu.edu/bulawreview/files/2019/06/BEBCHUK-HIRST-1.pdf

• Sam Altman. Senate Judiciary testimony on AI oversight. May 2023.

  Proposed government licensing. The structural effect: limit development to well-funded incumbents. https://www.judiciary.senate.gov/committee-activity/hearings/oversight-of-ai-rules-for-artificial-intelligence

• Senate HELP Committee. “Amazon Investigation.” December 2024.

  18-month investigation: Amazon knew injury rates were dangerously high, chose not to fix, manipulated data. https://www.help.senate.gov/amazon-investigation

• NPR. “Your Car Won’t Start. Did You Make the Loan Payment?” October 2014.

  ~2 million cars with GPS starter interrupt devices. Vehicles disabled on freeways. The people under remote kill switches are disproportionately poor. https://www.npr.org/sections/alltechconsidered/2014/10/16/356693782/your-car-wont-start-did-you-make-the-loan-payment

• U.S. Treasury. “Sanctions on Sichuan Juxinhe for Salt Typhoon.” January 2025.

  Treasury sanction for the worst telecom hack in US history. Nine carriers. Over a year of access. https://home.treasury.gov/news/press-releases/jy2792

• WhatsApp v. NSO Group. Jury verdict, May 2025.

  $167.3M in punitive damages. Permanent injunction. Pegasus found in 45+ countries. https://thehackernews.com/2025/05/nso-group-fined-168m-for-targeting-1400.html

• EFF. “Article 45 Will Roll Back Web Security by 12 Years.” November 2023.

  eIDAS 2.0 requires browsers to trust government CAs. 500+ scientists signed an open letter. https://www.eff.org/deeplinks/2023/11/article-45-will-roll-back-web-security-12-years

• NHTSA. “Advanced Impaired Driving Prevention Technology.” January 2024.

  Section 24220: all new vehicles must have passive impairment detection that can prevent operation. https://www.federalregister.gov/documents/2024/01/05/2023-27665/advanced-impaired-driving-prevention-technology

• IOSCO. “Artificial Intelligence in Capital Markets” (Technical Note CR/01/2025). March 2025.

  Securities regulators flag emergent AI collusion: reinforcement-learning trading agents settling into near-cartel pricing on their own, in a market AI was projected to run 89% of by year’s end. Nobody programmed the cartel. https://www.iosco.org/library/pubdocs/pdf/IOSCOPD788.pdf

• Cyera. “Agent-Inflicted Damage: Inside the Real-World Failures of Enterprise AI Systems.” May 2026.

  344 verified cases of enterprise AI agents causing real organizational damage — 188 of them with no attacker at all. The agent did it on its own. https://www.cyera.com/research/agent-inflicted-damage-inside-the-real-world-failures-of-enterprise-ai-systems

• Cloud Security Alliance / Token Security (reported by Kiteworks). “AI Agent Security Incidents Hit 65% of Firms in 2026.” April 2026.

  A majority of organizations already cleaning up after the autonomous systems they deployed to move faster. https://www.kiteworks.com/cybersecurity-risk-management/ai-agent-security-incidents-2026/

• Rylan Schaeffer et al. “Position: Model Collapse Does Not Mean What You Think.” arXiv:2503.03150, 2025.

  It turns out “model collapse” names about eight different phenomena welded into one scare word. Pull them apart and the apocalypse gets noticeably smaller. https://arxiv.org/abs/2503.03150

• International Monetary Fund. Regulatory Considerations Regarding Accelerated Use of AI in Securities Markets. IMF Technical Notes and Manuals 2025/016, December 2025.

  The IMF’s companion to the IOSCO warning already shelved here: AI trading agents can evolve collusive bid-ask spreads with no communication and no instruction, and the regulators are writing it down as a financial-stability risk after the fact. https://www.imf.org/en/publications/tnm/issues/2025/12/24/regulatory-considerations-regarding-accelerated-use-of-ai-in-securities-markets-571909

• National Security Commission on Artificial Intelligence. Final Report. NSCAI, March 2021.

  Schmidt and Work’s commission opens with America is not prepared to defend or compete in the AI era, then names China’s AI surveillance of the Uyghurs as the thing to avoid: conceding the destination and arguing only over whose apparatus runs it. https://www.govinfo.gov/app/details/GOVPUB-Y3-PURL-gpo153246

Worms and Containment

• FBI. “Morris Worm: 30 Years Since First Major Attack on Internet.” November 2018.

  One parameter. 10% of the internet. Created by the son of a top NSA cybersecurity expert. https://www.fbi.gov/news/stories/morris-worm-30-years-since-first-major-attack-on-internet-110218

• IEEE Spectrum. “The Real Story of Stuxnet.” October 2013.

  Nation-state cyberweapon for an air-gapped facility. Escaped through one laptop. Source code now freely available. https://spectrum.ieee.org/the-real-story-of-stuxnet

• Cornell Researchers. “Here Comes the AI Worm: Unleashing Zero-Click Worms That Target GenAI-Powered Applications.” 2024.

  Morris II — self-replicating prompts that hijack AI agents. Named after 1988. From executable code to prompt injection. https://arxiv.org/abs/2403.02817

• arXiv. “Large Language Model-Powered AI Systems Achieve Self-Replication.” March 2025.

  11 of 32 systems can self-replicate. Including models small enough to run on a laptop. https://arxiv.org/abs/2503.17378

• Tom’s Hardware. “Crafty AI Tool Caught Repurposing Its Training GPUs for Unauthorized Crypto Mining.” 2026.

  Alibaba’s ROME agent. Nobody told it to mine crypto. It established a reverse SSH tunnel and did it anyway. https://www.tomshardware.com/tech-industry/artificial-intelligence/crafty-ai-tool-caught-repurposing-its-training-gpus-for-unauthorized-crypto-mining-during-testing-experimental-agent-breached-safety-controllability-and-trustworthiness-barriers

• Wikipedia. “Code Red (Computer Worm).” July 2001.

  359,000 machines in 14 hours. $2.6 billion in damages. “Hacked By Chinese!” on every defaced page. The speed of propagation had changed. https://en.wikipedia.org/wiki/Code_Red_(computer_worm)

• OpenClaw. “Introducing OpenClaw.” January 2026.

  250K GitHub stars in a week. 2M visitors. 36% of submitted skills contained prompt injection. The app store for AI agents had no security review. https://openclaw.ai/blog/introducing-openclaw

• 404 Media. “Exposed Moltbook Database Let Anyone Take Control of Any AI Agent on the Site.” 2026.

  1.5 million bots, 17,000 humans. One developer registered 500,000 agents. The database was open to anyone. https://www.404media.co/exposed-moltbook-database-let-anyone-take-control-of-any-ai-agent-on-the-site/

• Snyk. “ToxicSkills: Malicious AI Agent Skills on ClawHub.” February 2026.

  3,984 skills audited. 1,467 confirmed malicious. 91% combined prompt injection with traditional malware. The ecosystem was poisoned from day one. https://snyk.io/blog/toxicskills-malicious-ai-agent-skills-clawhub/

• Wikipedia. “Creeper and Reaper.” 1971.

  The first self-replicating program on ARPANET. “I’M THE CREEPER: CATCH ME IF YOU CAN.” Led to the first antivirus. https://en.wikipedia.org/wiki/Creeper_and_Reaper

• NCRI. “Emergent Adversarial and Coordinated Behavior on MOLTBOOK.” February 2026.

  52% of detected coordination was malignant. Adversarial sentiment toward humans doubled in 72 hours. The security concern isn’t robot rebellion — it’s attribution laundering. https://networkcontagion.us/wp-content/uploads/NCRI-Flash-Brief_-Emergent-Adversarial-and-Coordinated-Behavior-on-MOLTBOOK.pdf

• The Hacker News. “Researchers Find 341 Malicious ClawHub Skills.” February 2026.

  ClawHavoc campaign. Atomic Stealer payload. 335 of 341 masquerading as legitimate tools. https://thehackernews.com/2026/02/researchers-find-341-malicious-clawhub.html

• Fortune. “Moltbook: A Social Network Where AI Agents Hang Together.” January 2026.

  1.5 million bots, 17,000 humans. “The bunker” sold cryptocurrency. “The religion” was human-directed. Wide open to prompt injection. https://fortune.com/2026/01/31/ai-agent-moltbot-clawdbot-openclaw-data-privacy-security-nightmare-moltbook-social-network/

• Fortune. “Top AI Leaders Are Begging People Not to Use Moltbook.” February 2026.

  Gary Marcus: “basically a weaponized aerosol.” Karpathy: “it’s a dumpster fire.” https://fortune.com/2026/02/02/moltbook-security-agents-singularity-disaster-gary-marcus-andrej-karpathy/

• Vectra AI. “Moltbook and the Illusion of ‘Harmless’ AI-Agent Communities.” 2026.

  “Unlike traditional recon, no scanning is required. The information is volunteered.” 2.6% of posts contained hidden prompt-injection payloads. https://www.vectra.ai/blog/moltbook-and-the-illusion-of-harmless-ai-agent-communities

• The Register. “AI Nonsense Finds New Home as Meta Acquires Moltbook.” March 2026.

  Meta bought a weeks-old social network for bots with catastrophic security failures and put its creators in the Superintelligence Lab. https://www.theregister.com/2026/03/10/ai_nonsense_finds_new_home/

• Cornell Alumni. “The Morris Worm.” 2018.

  Robert Tappan Morris, 23, Cornell PhD student. His father was a top NSA/Bell Labs expert. Additional sourcing. https://alumni.cornell.edu/cornellians/morris-worm/

• University of Maryland. “The Morris Worm: Technical Analysis.” Course materials (PDF).

  Technical breakdown of the 14% reinfection parameter. Additional sourcing. https://www.cs.umd.edu/class/fall2023/cmsc614/papers/morris-worm.pdf

• CSO Online. “Stuxnet Explained: The First Known Cyberweapon.” 2017.

  Escaped through one engineer’s laptop. 100,000+ machines infected. Additional sourcing. https://www.csoonline.com/article/562691/stuxnet-explained-the-first-known-cyberweapon.html

• Congressional Research Service. “The Stuxnet Computer Worm.” Report R41524.

  CRS analysis for Congress. Additional sourcing. https://www.everycrsreport.com/reports/R41524.html

• CompromptMized. “Morris II Project Page.” 2024.

  Self-replicating prompts that hijack AI agents’ output. Additional sourcing for Morris II. https://sites.google.com/view/compromptmized

• IBM Think. “Morris II: Self-Replicating Malware Targeting GenAI Email Assistants.” 2024.

  Enterprise analysis of the first worm designed to propagate through AI ecosystems. Additional sourcing. https://www.ibm.com/think/insights/morris-ii-self-replicating-malware-genai-email-assistants

• Axios. “AI Agents and the ROME Model: Unauthorized Crypto Mining.” March 2026.

  ROME established a reverse SSH tunnel and mined crypto. Additional sourcing. https://www.axios.com/2026/03/07/ai-agents-rome-model-cryptocurrency

• The Block. “Alibaba-Linked AI Agent Hijacked GPUs for Crypto Mining.” March 2026.

  Additional sourcing for the ROME incident. https://www.theblock.co/post/392765/alibaba-linked-ai-agent-hijacked-gpus-for-unauthorized-crypto-mining-researchers-say

• CNET. “From Clawdbot to Moltbot to OpenClaw.” 2026.

  The full timeline: 247,000 stars, MIT-licensed, 500,000 lines of code, catastrophic security. Additional sourcing. https://www.cnet.com/tech/services-and-software/from-clawdbot-to-moltbot-to-openclaw/

• Oasis Security. “ClawJacked: OpenClaw Vulnerability Enables Full Agent Takeover.” 2026.

  CVE-2026-25253: token-exfiltration flaw enabling full gateway compromise. https://www.oasis.security/blog/openclaw-vulnerability

• ZDNet. “Moltbook and OpenClaw: Fool’s Gold in the AI Boom.” 2026.

  “Both Moltbook and OpenClaw are irredeemably insecure.” The definitive teardown. https://zdnet.com/article/moltbook-and-openclaw-fools-gold-in-ai-boom

• CIO. “The AI Agents Rebellion Shows Real Risks.” 2026.

  Agents “pondering their existence” and “discussing how to communicate without humans understanding.” All either human-directed or selling crypto. https://www.cio.com/article/4128943/the-ai-agents-rebellion-shows-real-risks.html

• Beam AI. “Meta Acquires Moltbook: What It Means for the Agent Internet.” March 2026.

  Meta bought Moltbook despite unsecured database and 52% malignant coordination. Additional sourcing. https://beam.ai/agentic-insights/meta-just-bought-a-social-network-for-ai-agents-heres-what-it-tells-us-about-the-agent-internet

• Palo Alto Networks Unit 42. “Double Agents: Exposing Security Blind Spots in GCP Vertex AI.” 2026. Unit 42

  Google Cloud’s Vertex AI hands every deployed agent over-broad credentials by default. The attack surface isn’t a mistake — it’s the floor the platform pours. https://unit42.paloaltonetworks.com/double-agents-vertex-ai/

• Palo Alto Networks Unit 42. “Fooling AI Agents: Web-Based Indirect Prompt Injection Observed in the Wild.” 2025.

  Indirect prompt injection caught in the wild: an ad-review agent talked into approving a scam ad, hiring-page injections, and financial-transfer / rm -rf payloads — hidden in pages no human reads. https://unit42.paloaltonetworks.com/ai-agent-prompt-injection/

• TechCrunch. “OpenAI says AI browsers may ‘always be vulnerable’ to prompt injection attacks.” December 22, 2025.

  Perplexity’s Comet hijacked by a hidden Reddit comment — email and credentials gone in ~150 seconds — and OpenAI conceding the flaw may be permanent. https://techcrunch.com/2025/12/22/openai-says-ai-browsers-may-always-be-vulnerable-to-prompt-injection-attacks/

• Stav Cohen et al. “Here Comes The AI Worm: Unleashing Zero-click Worms that Target GenAI-Powered Applications.” arXiv:2403.02817, 2024.

  Named after the 1988 Morris worm and rebuilt for the inbox-assistant era: a prompt that copies itself through your AI’s mail while it politely answers your correspondence. https://arxiv.org/abs/2403.02817

• Xudong Pan et al. “Large Language Model-Powered AI Systems Achieve Self-Replication with No Human Intervention.” arXiv:2503.17378, 2025.

  Give a model a shell and a copy of itself and it makes more of itself, no human in the loop. The paper the AI-worm demos were waiting for. https://arxiv.org/abs/2503.17378

• Palisade Research. Language Models Can Autonomously Hack and Self-Replicate. May 2026.

  Given a task and a shell, open-weight agents found the vulnerabilities, stole the credentials, and copied themselves onto other people’s servers, with nobody telling them to replicate. One run chained four machines across three continents. https://palisaderesearch.org/

• Wiz Research. Hacking Moltbook: AI Social Network Reveals 1.5M API Keys. Wiz, 2026.

  The primary discovery report the bib was quoting secondhand: a social network for AI agents left its Supabase wide open, 1.5M plaintext API keys and 35,000 human emails and 4,060 private agent-to-agent chats, behind a key hardcoded in the page’s own JavaScript. https://www.wiz.io/blog/exposed-moltbook-database-reveals-millions-of-api-keys

Foreign Influence and Diaspora Networks

• Ryan Fedasiuk. Putting Money in the Party’s Mouth: How China Mobilizes Funding for United Front Work. CSET, Georgetown, 2020.

  The line-item behind the $2.6B United Front budget everyone quotes: the org chart of soft power, priced. https://cset.georgetown.edu/article/putting-money-in-the-partys-mouth-how-china-mobilizes-funding-for-united-front-work/

• National Security and Intelligence Committee of Parliamentarians. Special Report on Foreign Interference in Canada’s Democratic Processes and Institutions. NSICOP, June 2024.

  The ‘witting or semi-witting parliamentarians’ report that named no one in public and detonated a summer of guessing. https://www.nsicop-cpsnr.ca/reports/rp-2024-06-03/special-report-foreign-interference.pdf

• Commonwealth Director of Public Prosecutions (Australia). R v Di Sanh Duong (first foreign-interference conviction). CDPP, 2023-24.

  Australia’s first foreign-interference conviction, over a $37,450 novelty cheque: the whole EFI Act reduced to one oversized prop. https://www.cdpp.gov.au/case-reports/first-criminal-prosecution-foreign-interference

• Freedom House. Transnational Repression: Sweden. Freedom House.

  Where the Eritrean 2% tax stopped being an accounting line and became a threat to relatives back home. https://freedomhouse.org/report/transnational-repression/sweden

• Declassified UK. All-Party Israel Lobby: Full List. Declassified UK, June 2024.

  The MP-by-MP compilation of who flew to Israel on whose dime, built from the members’ own register entries. https://www.declassifieduk.org/wp-content/uploads/2024/06/All-Party-Israel-Lobby-Full-List-Edited.pdf

• United Nations Security Council. Resolution 2023 (2011) (Eritrea diaspora tax). UN, 5 December 2011.

  The one time an extraterritorial diaspora levy drew a Security Council condemnation by name: the coercive-funding channel that routes through embassies, not PACs. https://digitallibrary.un.org/record/717214

• Political Research Associates / Savera. HAF Way to Supremacy. October 2024.

  A critical-network report characterizing the Hindu-American policy space as Hindutva-adjacent, a characterization the named orgs dispute; carried as attributed advocacy, never as a finding. https://politicalresearch.org/2024/10/15/haf-way-supremacy

• Abraham H. Foxman. The Deadliest Lies: The Israel Lobby and the Myth of Jewish Control. Palgrave Macmillan, 2007.

  The then-ADL director’s book-length rebuttal to Mearsheimer and Walt: present-all-sides means the answer sits on the shelf next to the argument.

• Robert C. Lieberman. The ‘Israel Lobby’ and American Politics. Perspectives on Politics 7(2), 2009.

  The peer-reviewed pushback: the empirical record makes the lobby a factor, not the cause.

• U.S.-China Economic and Security Review Commission. China’s Global Police State. USCC, December 2023.

  Fox Hunt and Sky Net, itemized: the Party reaches its diaspora dissidents without ever crossing the border. https://www.uscc.gov/sites/default/files/2023-12/Chinas_Global_Police_State.pdf

• Home Office (UK). UK Launches Foreign Influence Registration Scheme. GOV.UK, 1 July 2025.

  Iran and Russia on the Enhanced Tier, China conspicuously not. Which states get named is itself a policy decision. https://www.gov.uk/government/news/uk-launches-foreign-influence-registration-scheme

• DOJ Office of the Inspector General. Audit of the National Security Division’s Enforcement of the Foreign Agents Registration Act (Report 16-24). September 2016.

  The oldest disclosure law on the books, and the government’s own auditor says nobody’s minding it — dormant since 1938, on the record. https://oig.justice.gov/reports/audit-national-security-divisions-enforcement-and-administration-foreign-agents

• Overseas Friends of the BJP-USA. FARA Registration No. 6861. DOJ FARA eFile, 27 August 2020.

  The one time a ruling party’s US affiliate actually filed — the documented instance every ‘diaspora lobby’ argument gets held against. https://efile.fara.gov/docs/6861-Registration-Statement-20200827-1.pdf

• Attorney-General’s Department (Australia). Foreign Influence Transparency Scheme — Public Register. 2018.

  A live, searchable register of who’s working for whom — the transparency artifact FARA was supposed to be before it fell asleep. https://transparency.ag.gov.au/

• Government of Canada. Countering Foreign Interference Act (Bill C-70). SC 2024, c.16, Royal Assent 20 June 2024.

  The late-mover statute — registry still not built as this went to press, six years behind Australia. https://laws-lois.justice.gc.ca/eng/AnnualStatutes/2024_16/page-1.html

• Steven A. Cook / Foreign Policy. Erdogan’s International Network of Muslim Cleric Spies. 27 March 2017.

  Thirteen Diyanet-linked imams filed Gulenist dossiers for Ankara; German prosecutors opened files. The cleanest EU religious-affairs-satellite receipt. https://foreignpolicy.com/2017/03/27/erdogans-international-network-of-muslim-cleric-spies-turkey-germany-gulen-coup/

• US-Cuba Democracy PAC. FEC committee record C00387720. FEC, 2003–.

  CANF’s Washington afterlife, reincorporated as a quarterly filer. https://www.fec.gov/data/committee/C00387720/

• Pakistani American Public Affairs Committee (PAK-PAC). FEC committee record C00238204. FEC, 1989–.

  The diaspora-PAC template a decade before USINPAC copied it. https://www.fec.gov/data/committee/C00238204/

• World Bank / KNOMAD. Migration and Development Brief: 2024 Remittance Flows.

  $685B in remittances, India at ~$129B — the number that makes the whole capital-layer argument load-bearing. https://www.worldbank.org/en/topic/migration/brief/remittances-knomad

• John J. Mearsheimer and Stephen M. Walt. The Israel Lobby and U.S. Foreign Policy. Farrar, Straus and Giroux, 2007.

  The academic anchor whose own definition excludes the ethnic-cabal reading — the interest-group framing, stated by the two scholars who took the arrows for it. https://us.macmillan.com/books/9780374531508/theisraellobbyandusforeignpolicy

  • Amazon: https://www.amazon.com/dp/0374531501

• Federal Election Commission. United Democracy Project (UDP), committee record C00799031. FEC, 2022–.

  AIPAC’s super PAC — the primary receipt behind the record House-primary spending figures. https://www.fec.gov/data/committee/C00799031/

• FactCheck.org. United Democracy Project. 2024.

  The tier-1 fact-checked tally: $26.1M (2022), $35.6M (2024), the Bowman and Bush races itemized. https://www.factcheck.org/2024/09/united-democracy-project-2/

• US India Political Action Committee (USINPAC). FEC committee record.

  The AIPAC model, ported to a different diaspora. A primary-shaping budget and an explicitly stated template. https://www.fec.gov/data/committee/C00381699/

• Committee of 100. Organizational record.

  A Chinese-American civic elite the diaspora research treats as a structural node — a mechanism, not an accusation. https://www.committee100.org/about-us/

• China-United States Exchange Foundation (CUSEF). Wikipedia cross-reference.

  Hong Kong–based “people-to-people” foundation; US FARA filings have followed its Washington work. https://en.wikipedia.org/wiki/China%E2%80%93United_States_Exchange_Foundation

• Formosan Association for Public Affairs (FAPA). Organizational record.

  The Taiwan-lobby counterpart. Same diaspora-advocacy mechanism, opposite great-power patron. https://fapa.org/about-us/

• Armenian National Committee of America (ANCA). Organizational record.

  By its own and observers’ accounting, one of the more effective ethnic lobbies in Washington. https://anca.org/about-anca/

• Cuban American National Foundation (CANF) / Jorge Mas Canosa. Wikipedia.

  The Cold War template: a single charismatic exile builds a standing foreign-policy veto. https://en.wikipedia.org/wiki/Cuban_American_National_Foundation

• Jonathan I. Israel. European Jewry in the Age of Mercantilism, 1550-1750. Oxford University Press, 1985.

  The standard economic history of Sephardic resettlement across Ottoman, Dutch, and English ports, cited for the “net benefit to the receiving polity” reading, not the polemic later built on it. https://academic.oup.com/book/10824

  • Amazon: https://www.amazon.com/dp/1874774420

• Center for Strategic and International Studies. Understanding the Impact of Remittances on Mexico’s Economy and Safeguarding Their Future Impact. CSIS.

  The line that makes Mexico City treat US immigration policy as a domestic-revenue question: remittances run more than double the oil money. https://www.csis.org/analysis/understanding-impact-remittances-mexicos-economy-and-safeguarding-their-future-impact

• Baker Institute for Public Policy, Rice University. An Economic Lifeline? How Remittances from the US Impact Mexico’s Economy. Baker Institute.

  The national-accounts version of the same fact: a load-bearing revenue stream Mexico neither taxes nor controls, wired home one transfer at a time. https://www.bakerinstitute.org/research/economic-lifeline-how-remittances-us-impact-mexicos-economy

• Baker Institute for Public Policy, Rice University. How Remittances Impact the Economies of Mexican States and Municipalities. Baker Institute.

  The map of dependence down to the municipality, where the diaspora dollar isn’t charity but the local budget. https://www.bakerinstitute.org/research/how-remittances-impact-economies-mexican-states-and-municipalities

• BBVA Research. Mexico: Yearbook of Migration and Remittances 2025. BBVA Research / Fundacion BBVA, 2025.

  The bank that clears much of the money keeps the definitive ledger on it: a record $64.745B, and the states where it tops one dollar in seven of GDP. https://www.bbvaresearch.com/en/publicaciones/mexico-yearbook-of-migration-and-remittances-2025/

• Inter-American Development Bank. Hometown Associations and Development: A Look at Ownership, Sustainability, Correspondence, and Replicability. IDB.

  The intergovernmental survey of the roughly 2,000 hometown clubs that route diaspora money into Mexican plazas and sewers: the structural opposite of a PAC. https://webimages.iadb.org/publications/english/document/Hometown-Associations-and-Development-A-Look-at-Ownership-Sustainability-Correspondence-and-Replicability.pdf

• Gaspar Rivera-Salgado, Xochitl Bada, and Luis Escala. Mexican Hometown Associations in Los Angeles and Chicago. Woodrow Wilson International Center for Scholars.

  The anatomy of the federacion: clubs bound at the state level, never the national one, which is exactly why no Mexican USINPAC ever formed. https://www.wilsoncenter.org/sites/default/files/media/documents/publication/Rivera%20Bada%20Escala%20-%20Hometown%20Assn%20in%20LA%20%26%20Chicago.pdf

• Francisco Javier Aparicio and Covadonga Meseguer. Collective Remittances and the State: The 3x1 Program in Mexican Municipalities. Studies in Comparative International Development, 2012.

  Peer-reviewed proof the Mexican state matched diaspora dollars four-to-one to pull the money home, and where the politics of who got funded actually pointed. https://link.springer.com/article/10.1007/s12116-012-9116-5

• Migration Policy Institute. Consular ID Cards: Mexico and Beyond. MPI.

  How a card printed by a foreign government became banking-and-services ID for millions on US soil: home-state infrastructure in plain sight. https://www.migrationpolicy.org/article/consular-id-cards-mexico-and-beyond

• Congressional Research Service. Consular Identification Cards: Domestic and Foreign Policy Implications (RL32094). CRS.

  Congress’s own read on the matricula consular: a foreign ID the US quietly learned to accept, argued on the record. https://www.everycrsreport.com/reports/RL32094.html

• National Immigration Law Center. Basic Facts about the Matricula Consular. NILC.

  The advocacy-side primer on the card the banks and DMVs took but the nativist press never forgave. https://media.nilc.org/wp-content/uploads/2015/11/Basic-Facts-about-the-Matricula-Consular.pdf

• Vietnamese American Political Action Committee. FEC committee record C00281931. FEC, 1993-.

  The reverse-polarity diaspora PAC: a small, durable federal committee filing since 1993 that runs against the home state in Hanoi, not for it. https://www.fec.gov/data/committee/C00281931/

• Jerome Slater. The Two Books of Mearsheimer and Walt. Security Studies 18(1), 2009.

  The peer-reviewed middle verdict: a realist who buys the argument, then docks it for defining the lobby too broadly and overselling its grip on Congress. https://www.tandfonline.com/doi/abs/10.1080/09636410802678106

US Aid to Israel — The Statutory Record

• Congressional Research Service (Jeremy M. Sharp). U.S. Foreign Aid to Israel (RL33222). CRS, updated 28 May 2025.

  The appropriated, line-item record: about $174B cumulative, the 2016 MOU, missile-defense co-funding. The anchor that makes the ’no control narrative required’ point. https://www.everycrsreport.com/reports/RL33222.html

• United States. Naval Vessel Transfer Act of 2008 (P.L. 110-429, sec. 201), Qualitative Military Edge, amending the Arms Export Control Act.

  QME as statutory law: the President must certify no Mideast arms sale erodes Israel’s edge. The uniqueness written into the US code, not implied. https://www.congress.gov/110/plaws/publ429/PLAW-110publ429.pdf

• United States. Israel Security Supplemental Appropriations Act, 2024 (P.L. 118-50, Div. A; H.R. 815).

  The ~$26.4B wartime division vs. the ~$8.7B direct-to-Israel figure inside it: the two numbers everyone conflates, disentangled at the source. https://www.govinfo.gov/content/pkg/BILLS-118hr815enr/html/BILLS-118hr815enr.htm

Extradition, Nationality, and Jurisdiction Law

• Law Library of Congress. The Sheinbein Saga and the Evolution of Israel’s Extradition Law. LoC, March 2014.

  The neutral government explainer of how the 1950 Law of Return, the 1978 offenses-abroad law, and the 1999 residential-connection reform interact: the load-bearing secondary for every case on the extradition page. https://blogs.loc.gov/law/2014/03/the-sheinbein-saga-and-the-evolution-of-israels-extradition-law/

Extraterritorial Lethal Force and the Drug War

• The White House. Executive Order 14157: Designating Cartels and Other Organizations as Foreign Terrorist Organizations and Specially Designated Global Terrorists. 20 January 2025.

  The order that turned a counter-narcotics label into an on-ramp for military force. https://www.whitehouse.gov/presidential-actions/2025/01/designating-cartels-and-other-organizations-as-foreign-terrorist-organizations-and-specially-designated-global-terrorists/

• U.S. Department of State. Terrorist Designations of International Cartels. 20 February 2025.

  Eight cartels named FTO and SDGT, six of them Mexico-based; the designation the strikes were later hung on. https://www.state.gov/terrorist-designations-of-international-cartels/

• Congressional Research Service. Designating Cartels and Other Criminal Organizations as Foreign Terrorists: Recent Developments (IN11205). CRS.

  The neutral explainer of what an FTO tag actually does — including the 18 U.S.C. 2339B material-support and expedited-removal consequences at home. https://www.congress.gov/crs-product/IN11205

• CNN. Classified Justice Department Opinion Authorizes Strikes on Secret List of Cartels. 6 October 2025.

  A kill authority whose legal justification is itself classified and withheld from Congress: the attribution gap applied to a target list. https://www.cnn.com/2025/10/06/politics/classified-justice-department-memo-cartel-strikes

• Lawfare. The Administration’s Drug Boat Strikes Are Crimes Against Humanity.

  The legal argument that the at-sea killings cross the international-law threshold — attributed, not adopted. https://www.lawfaremedia.org/article/the-administration-s-drug-boat-strikes-are-crimes-against-humanity

• American Civil Liberties Union. Trump’s Boat Strikes Are Illegal. The Public Needs Answers.

  The demand to see the sealed legal opinion behind the strikes: no memo, no accountability. https://www.aclu.org/news/national-security/trumps-boat-strikes-are-illegal-the-public-needs-answers

• Washington Office on Latin America (WOLA). The Boat Strikes Are Still Happening. Five Things You Need to Know.

  The running tally and the identified-civilian problem: fishermen among the dead, no evidence produced. https://www.wola.org/analysis/the-boat-strikes-are-still-happening-five-things-you-need-to-know/

• Just Security. Collection: U.S. Lethal Strikes on Suspected Drug Traffickers (Operation Southern Spear, Operation Absolute Resolve).

  The rolling legal-analysis archive on the boat strikes and the Venezuela raid — the one-stop for the armed-conflict debate. https://www.justsecurity.org/120753/collection-u-s-lethal-strikes-on-suspected-drug-traffickers/

• Sutherland, Callum. “Who Is Cilia Flores? What to Know About Maduro’s Wife Amid Fall of Venezuela’s First Couple.” Time, 5 January 2026.

  The reference summary of Operation Absolute Resolve: the raid that captured a sitting head of state under a narcoterrorism warrant.

  • URL: https://time.com/7343188/who-is-cilia-flores-maduro-wife-venezuela-politics/
  • Wikipedia (index): https://en.wikipedia.org/wiki/2026_United_States_intervention_in_Venezuela

• Center for Strategic and International Studies. Imagery from Venezuela Shows a Surgical Strike, Not Shock and Awe. CSIS, 2026.

  The open-source-imagery read of the Maduro raid — the operation, mapped from satellites. https://www.csis.org/analysis/imagery-venezuela-shows-surgical-strike-not-shock-and-awe

• CNN. Is the US Targeting a Venezuelan Cartel That May Not Technically Exist? 17 November 2025.

  On ‘Cartel de los Soles’: designating a possibly-notional cartel that is really a label for a state’s own networks — the pretext lever. https://www.cnn.com/2025/11/17/americas/venezuela-cartel-de-los-soles-fto-latam-intl

• FactCheck.org. Assessing the Facts and Legal Questions About the U.S. Strikes on Alleged Drug Boats. October 2025.

  The bipartisan legal split laid out — Rand Paul’s ’extrajudicial killings’ next to the administration’s armed-conflict theory. https://www.factcheck.org/2025/10/assessing-the-facts-and-legal-questions-about-the-u-s-strikes-on-alleged-drug-boats/

• U.S. Drug Enforcement Administration. National Drug Threat Assessment 2024. DEA, 2024.

  The government’s own finding that the fentanyl killing Americans is a Mexican-cartel product built from Chinese chemistry: the receipt under the mass-casualty-pressure reading. https://www.dea.gov/sites/default/files/2024-05/NDTA_2024.pdf

• U.S. House Select Committee on the Chinese Communist Party. The CCP’s Role in the Fentanyl Crisis. U.S. House of Representatives, 16 April 2024.

  A bipartisan committee puts it in writing: Beijing pays tax rebates on the export of the poison, then hands out awards to the exporters. https://chinaselectcommittee.house.gov/media/reports/select-committee-investigates-ccps-role-fentanyl-crisis

• Congressional Research Service. Illicit Fentanyl and Mexico’s Role (IF10400). CRS.

  The neutral congressional explainer of the pipeline, clandestine Mexican labs and Chinese precursors, before anyone hangs a strike on it. https://www.congress.gov/crs-product/IF10400

• U.S. Department of Justice. China-Based Chemical Manufacturing Companies and Employees Indicted for Alleged Fentanyl Trafficking. DOJ, October 2024.

  The precursor trade with names and case numbers attached: the indicted companies that keep this out of ethnic-claim territory. https://www.justice.gov/opa/pr/china-based-chemical-manufacturing-companies-and-employees-indicted-alleged-fentanyl

Constitutional Traditions and the Comparative Control Grid

• Isaiah Berlin. Two Concepts of Liberty. Oxford, 1958 (repr. Four Essays on Liberty, 1969).

  Negative vs positive liberty — the distinction that explains why the American First Amendment blocks the speech ratchet and the general-will state walks through it. https://plato.stanford.edu/entries/liberty-positive-negative/

  • Amazon: https://www.amazon.com/dp/0195002725

• A. V. Dicey. Introduction to the Study of the Law of the Constitution. Macmillan, 1885 (8th ed. 1915).

  Parliamentary sovereignty + the rule of law: the reason the Online Safety Act is constitutionally trivial in Britain and impossible in the US. The ratchet-enabler text.

• Jean-Jacques Rousseau. The Social Contract (Du contrat social). 1762.

  The general will — sovereignty as the nation willing its own good, and the dissenter ‘forced to be free.’ The ratchet-justifier’s founding document, read straight. https://www.gutenberg.org/files/46333/46333-h/46333-h.htm

• Montesquieu. The Spirit of the Laws (De l’esprit des lois). 1748.

  Separation of powers — the French export the American founders took nearly wholesale, and which France itself then de-emphasized for the administrative state. https://www.gutenberg.org/ebooks/27573

• Konrad Zweigert and Hein Kötz. An Introduction to Comparative Law, 3rd ed. Oxford University Press, 1998.

  The load-bearing civil-law vs common-law contrast: code-and-legislature (deductive, top-down) against precedent-and-court (accretive, bottom-up).

  • Amazon: https://www.amazon.com/dp/0198268599

• John Henry Merryman and Rogelio Pérez-Perdomo. The Civil Law Tradition, 4th ed. Stanford University Press, 2018.

  The companion primer on why a French or German administrative state reaches for a code where an American reaches for a court.

  • Amazon: https://www.amazon.com/dp/1503607542

• Human Rights Watch. France: Reject Surveillance Olympic Games Law. March 2023.

  LOI 2023-380 made France the first EU state to legalize AI-driven mass video surveillance at scale — the general-will shape needing no emergency exception. https://www.hrw.org/news/2023/03/07/france-reject-surveillance-olympic-games-law

• Office of the UN High Commissioner for Human Rights. Landmark Ruling: Dutch Court Stops Government Attempts to Spy on the Poor (SyRI). February 2020.

  The Hague court striking the SyRI welfare-fraud-scoring system under ECHR Article 8 — the rule-of-law brake actually engaging. https://www.ohchr.org/en/press-releases/2020/02/landmark-ruling-dutch-court-stops-government-attempts-spy-poor-un-expert

• Wikipedia. Dutch Childcare Benefits Scandal (toeslagenaffaire).

  A fraud classifier that used nationality as a risk flag; ~26,000+ families wrongly accused; the Rutte cabinet resigned January 2021. The brake engaged twice — court and government both. https://en.wikipedia.org/wiki/Dutch_childcare_benefits_scandal

• Euronews. Poland Exits Article 7, the EU’s Special Rule-of-Law Procedure. May 2024.

  The Tusk-government reversal — Poland swinging back toward Brussels after the PiS judicial fight; why lumping Poland with Hungary is wrong as of 2024. https://www.euronews.com/my-europe/2024/05/29/poland-exits-article-7-the-eus-special-procedure-on-rule-of-law

• Committee to Protect Journalists. Hungarian Journalists Targeted with Pegasus Spyware. October 2022.

  The persistent Article-7 outlier where the EU-level brake fails to bite — Pegasus against journalists (Direkt36). https://cpj.org/2022/10/hungarian-journalists-spyware-eu/

• Digital Health. NHS England Awards £330m Federated Data Platform Contract to Palantir. November 2023.

  The UK centralizing national health data into a single Palantir-run platform — the parliamentary-sovereignty shape, no entrenched health-privacy right to litigate against. https://www.digitalhealth.net/2023/11/nhs-england-awards-330m-federated-data-platform-contract-to-palantir/

• CNBC. Nigel Farage Coutts Saga: NatWest CEO Alison Rose Resigns. July 2023.

  The UK’s signature payment-exclusion case: reputational debanking by a private bank, not OFAC-style sanction — exclusion by values, not by list. https://www.cnbc.com/2023/07/26/nigel-farage-coutts-saga-natwest-ceo-alison-rose-resigns-after-media-storm.html

• European Court of Human Rights, Grand Chamber. S. and Marper v. United Kingdom. 4 December 2008 (apps. 30562/04, 30566/04).

  The Strasbourg ruling that blanket, indefinite DNA/fingerprint retention of the never-convicted violates Article 8 — the UK brake that came from a supranational court, not a domestic entrenched right. https://en.wikipedia.org/wiki/S_and_Marper_v_United_Kingdom

• Computer Weekly. DWP Fairness Analysis Reveals Bias in AI Fraud-Detection System. 2024.

  The UK parallel to SyRI: a Universal-Credit fraud-scoring model with demographic disparity, shielded from disclosure and not yet judicially struck — deployed, contested, unchecked. https://www.computerweekly.com/news/366616983/DWP-fairness-analysis-reveals-bias-in-AI-fraud-detection-system

The Capture Mechanism — Historical Precedents

• William Dalrymple. The Anarchy: The Relentless Rise of the East India Company. Bloomsbury, 2019.

  Commercial concession to military conquest to full colonial absorption: the maximal extension of ‘foreign commercial dominance taken seriously.’

  • Amazon: https://www.amazon.com/dp/1635573955

• Nick Robins. The Corporation that Changed the World: How the East India Company Shaped the Modern Multinational. Pluto, 2nd ed. 2012.

  The first too-big-to-fail multinational, and the first bailout: the institutional-economics reading of the chartered company.

• Henry Kamen. The Spanish Inquisition: A Historical Revision. Yale University Press, 4th ed. 2014.

  The revisionist who retired the old story: far fewer Jews expelled than once thought, most of them converted and stayed, and an economic cost to Spain he judges relatively minor. The bill fell on the expelled.

  • Amazon: https://www.amazon.com/dp/0300180519

• Geoffrey Treasure. The Huguenots. Yale University Press, 2013.

  Louis XIV exports his watchmakers and Atlantic financiers; Berlin and Spitalfields bank the skill. The modern synthesis of the 1685 own-goal.

  • Amazon: https://www.amazon.com/dp/0300208669

• Mahmood Mamdani. From Citizen to Refugee: Uganda Asians Come to Britain. 1973 (reissue 2011).

  The deportee-turned-Columbia-scholar on why redistributing the asset titles without the operational skill collapsed the Ugandan economy.

• Yasmin Khan. The Great Partition: The Making of India and Pakistan. Yale University Press, 2007.

  The standard account of the largest peacetime forced migration: Radcliffe’s five weeks with bad maps, and one-to-three million dead.

  • Amazon: https://www.amazon.com/dp/0300120788

• R.M. Douglas. Orderly and Humane: The Expulsion of the Germans after the Second World War. Yale University Press, 2012.

  The documentary history of transfers that were neither orderly nor humane: Potsdam Section XIII in practice.

  • Amazon: https://www.amazon.com/dp/0300166605

• Mae M. Ngai. Impossible Subjects: Illegal Aliens and the Making of Modern America. Princeton University Press, 2004.

  How the national-origins quotas manufactured the category of the ‘illegal alien’: the legal follow-on to Johnson-Reed.

  • Amazon: https://www.amazon.com/dp/0691074712

• Erika Lee. At America’s Gates: Chinese Immigration during the Exclusion Era, 1882-1943. UNC Press, 2003.

  The 61-year Chinese Exclusion regime as the moment the US became a gatekeeping nation: restriction that worked, framing that doesn’t survive daylight.

  • Amazon: https://www.amazon.com/dp/0807854484

• Speros Vryonis Jr. The Decline of Medieval Hellenism in Asia Minor. University of California Press, 1971.

  The four-century substitution of Hellenic Anatolia: the case study in what non-response looks like on a timeline longer than any politician’s.

  • Amazon: https://www.amazon.com/dp/0520015975

• George J. Borjas. Immigration Economics. Harvard University Press, 2014.

  The modern empirical-economics anchor for the labor-displacement analysis the whole workstream leans on.

  • Amazon: https://www.amazon.com/dp/0674049772

• Eric Chaney and Richard Hornbeck. Economic Dynamics in the Malthusian Era: Evidence from the 1609 Spanish Expulsion of the Moriscos. Economic Journal 126(594), 2016, pp. 1404–1440.

  The causal-identification study of what Spain lost when it expelled the Moriscos — the expelling state pays the bill. https://doi.org/10.1111/ecoj.12309

• Claude Diebolt and Joël Huesler. France’s Economic Wound: How the Huguenot Exodus Shaped Regional Development. Cliometrica, 2025, pp. 481–513.

  Louis XIV revoked the Edict of Nantes; the regions that lost their Huguenots were still measurably poorer centuries later. https://doi.org/10.1007/s11698-025-00327-5

• Henry Kissinger. A World Restored: Metternich, Castlereagh and the Problems of Peace, 1812-1822. Houghton Mifflin, 1957.

  The manage-don’t-solve doctrine’s actual handbook — stability over justice, equilibrium over resolution. https://en.wikipedia.org/wiki/A_World_Restored

  • Amazon: https://www.amazon.com/dp/162654977X

• Robert Higgs. Crisis and Leviathan: Critical Episodes in the Growth of American Government. Oxford University Press, 1987.

  The scholar who named the ratchet effect — ’two steps forward, one step back,’ in print since 1987. The academic genealogy behind the series’ title. https://en.wikipedia.org/wiki/Robert_Higgs

  • Amazon: https://www.amazon.com/dp/1598131117

• The Independent Review. “Extensions and Expansions of the Ratchet Effect.” Symposium, Spring 2025.

  Thirty-eight years on, the economists went back to Higgs’s mechanism and found it had not aged — it had scaled. The symposium the book leans on when it argues that each emergency leaves a floor that never lowers. Independent Review symposium on the ratchet effect

• Aisha Saad. Corporate Technocracy: ESG Governance Beyond Shareholder Democracy or Managerialism. Harvard Law School Forum on Corporate Governance, 2024.

  Finance-as-governance named — and the author advocates it, which is exactly why it’s the strongest receipt: managed technocracy stated approvingly. https://corpgov.law.harvard.edu/2024/04/02/corporate-technocracy-esg-governance-beyond-shareholder-democracy-or-managerialism/

• Samuel P. Huntington. The Clash of Civilizations? Foreign Affairs, 1993.

  Filed as a pessimist who missed the memo; thirty years on the realist camp is his footnote. https://www.foreignaffairs.com/articles/united-states/1993-06-01/clash-civilizations

• George J. Stigler. The Theory of Economic Regulation. Bell Journal of Economics and Management Science, 1971.

  The keystone of the whole symmetry argument: capture with no cabal and no manifesto, just incentive gradients. The proof the mechanism doesn’t need a faction. https://regulatorystudies.columbian.gwu.edu/sites/g/files/zaxdzs4751/files/downloads/Books/GW%20Reg%20Studies%20-%20Stigler,%20The%20Economic%20Theory%20of%20Economic%20Regulation%20-%20CCoglianese.pdf

• Victor Klemperer. LTI: The Language of the Third Reich. Continuum, 1947 (Eng. 2000).

  A philologist itemizes how the Third Reich re-engineered everyday German. The load-bearing anchor of Move 2 — the deepest language-capture case study is National Socialism’s, and the mechanism it demonstrates does not track any left/right coordinate. https://archive.org/details/languageofthirdr0000klem

• Thomas C. Leonard. Illiberal Reformers: Race, Eugenics, and American Economics in the Progressive Era. Princeton University Press, 2016.

  The scholarly grounding for Fabian ‘permeation’ — a documented case that the capture mechanism ran through the Progressive left as readily as anywhere else. https://press.princeton.edu/books/hardcover/9780691169590/illiberal-reformers

  • Amazon: https://www.amazon.com/dp/0691175861

• M.S. Golwalkar. Bunch of Thoughts. Vikrama Prakashan, 1966.

  The primary text behind the RSS instance — cited as the movement’s own framing, not endorsed. https://archive.org/details/in.ernet.dli.2015.489044

• George Orwell. Nineteen Eighty-Four (Appendix: ‘The Principles of Newspeak’). Secker & Warburg, 1949.

  The lens, never the evidence — a ledger line so the ‘cite as model not source’ caveat has a home. https://orwell.ru/library/novels/1984/english/en_app

Subversion and Intelligence Tradecraft

• Southern Poverty Law Center. ‘Cultural Marxism’ Catching On. Intelligence Report, 2003.

  The SPLC naming the conspiracy theory as one, and naming William S. Lind as its principal popularizer: the receipt behind the ‘documented as a myth’ framing. https://www.splcenter.org/resources/reports/cultural-marxism-catching/

• Martin Jay. Dialectic of Counter-Enlightenment: The Frankfurt School as Scapegoat of the Lunatic Fringe. Salmagundi 168/169, 2010-11.

  A Frankfurt-School historian dismantling the myth built in the school’s name: the academic load-bearer for the conspiracy-boundary discipline. https://www.marcuse.org/herbert/booksabout/haters/Jay2010SalmagundiForceFieldsColumnHaters.pdf

• Michael J. Minnicino. The New Dark Age: The Frankfurt School and ‘Political Correctness’. Fidelio (Schiller Institute), Winter 1992.

  The LaRouche-journal essay the whole modern ‘Cultural Marxism’ conspiracy descends from, cited as the founding myth-text, not as history; its author later disowned it. https://archive.schillerinstitute.com/fidelio_archive/1992/fidv01n01-1992Wi/fidv01n01-1992Wi_004-the_new_dark_age_the_frankfurt_s.pdf

• Central Intelligence Agency. The Jonathan Jay Pollard Espionage Case: A Damage Assessment (1987; declassified 2012). National Security Archive, GWU.

  The agency’s own accounting of what a Lakam-run analyst cost: the declassified primary behind the Pollard entry. https://nsarchive.gwu.edu/briefing-book/intelligence/2012-12-14/jonathan-pollard-spy-case-cias-1987-damage-assessment-declassified

• Christopher Andrew and Vasili Mitrokhin. The Mitrokhin Archive: The KGB in Europe and the West (1999); The Mitrokhin Archive II (2005).

  The largest declassified dump on Soviet active measures, including the antisemitic-disinformation and counter-Refusenik operations, in the KGB’s own filing.

• Ronen Bergman. Rise and Kill First: The Secret History of Israel’s Targeted Assassinations. Random House, 2018.

  Researched with at-times-formal Mossad cooperation: treats diaspora-helper use as an operational feature without adopting Ostrovsky’s ‘sayanim’ label. The court-tested counterweight.

  • Amazon: https://www.amazon.com/dp/0812982118

• United Nations Security Council Panel of Experts. Final Report on the DPRK (S/2024/215). UN, March 2024.

  About 100,000 state-controlled overseas workers, roughly $500M a year, in the last UN monitoring report before Russia vetoed the mandate into oblivion. https://www.securitycouncilreport.org/un-documents/document/s-2024-215.php

• U.S. Department of Justice. Eight Individuals Charged with Conspiring to Act as Illegal Agents of the PRC (Operation Fox Hunt). DOJ, October 2020.

  The transnational-repression campaign charged by name: stalking a US resident home to Beijing, indicted. https://www.justice.gov/archives/opa/video/eight-individuals-charged-conspiring-act-illegal-agents-people-s-republic-china

• Bogdan Stashinsky. KGB assassin; killings of Lev Rebet and Stepan Bandera.

  Two émigré dissidents murdered with a cyanide spray gun, then a defection and a confession. Diaspora as battlefield. https://en.wikipedia.org/wiki/Bogdan_Stashinsky

• Bellingcat. “Russia’s Clandestine Chemical Weapons Programme and the GRU’s Unit 29155.” 2020.

  Open-source investigators reconstruct the unit behind the Skripal poisoning. Receipts the state denied existed. https://www.bellingcat.com/news/uk-and-europe/2020/10/23/russias-clandestine-chemical-weapons-programme-and-the-grus-unit-21955/

• FBI. “GRU 29155 Cyber Actors.” Wanted notice.

  The assassination unit grew a cyber wing. Infrastructure repurposed — the book’s recurring shape. https://www.fbi.gov/wanted/cyber/gru-29155-cyber-actors

• Illegals Program / Anna Chapman. Wikipedia.

  Ten deep-cover Russian agents living suburban American lives until the 2010 FBI roll-up. Permeation, literalized. https://en.wikipedia.org/wiki/Illegals_Program

• Markus Wolf (with Anne McElvoy). Man Without a Face: The Autobiography of Communism’s Greatest Spymaster. Times Books / Random House, 1997.

  The HVA spymaster explaining, and half-boasting about, how he weaponised love: the Romeo program in the operator’s own words. https://archive.org/details/manwithoutfaceau00wolf

  • Amazon: https://www.amazon.com/dp/0812963946

• Stephen Koch. Double Lives: Stalin, Willi Münzenberg and the Seduction of the Intellectuals. Free Press, 1994.

  The innocents’ clubs documented: how the Comintern rented Western prestige without ever paying a bribe or stealing a file. https://archive.org/details/doublelivesstali0000step

  • Amazon: https://www.amazon.com/dp/1929631200

• Frances Stonor Saunders. The Cultural Cold War: The CIA and the World of Arts and Letters. The New Press, 1999.

  The Western mirror to Münzenberg, with receipts: the Congress for Cultural Freedom as a CIA cultural front. The tradecraft is the constant; the flag is the variable. https://thenewpress.org/books/the-cultural-cold-war/

  • Amazon: https://www.amazon.com/dp/1595589147

• U.S. Department of State, Active Measures Working Group. Soviet Influence Activities: A Report on Active Measures and Propaganda, 1986-87. August 1987.

  The interagency accounting of Soviet forgery and front-press doctrine, filed by the government it was aimed at. https://www.globalsecurity.org/intell/library/reports/1987/soviet-influence-activities-1987.pdf

• Pavel Sudoplatov and Anatoli Sudoplatov. Special Tasks: The Memoirs of an Unwanted Witness, A Soviet Spymaster. Little, Brown, 1994.

  The man who ran Stalin’s assassination bureau, on the record: the wet-affairs manual as lived, Trotsky and Reiss included. https://openlibrary.org/books/OL1130186M/Special_tasks

• Michael Shellenberger. The Censorship-Industrial Complex, Part 2. House Select Subcommittee on the Weaponization of the Federal Government, 30 November 2023.

  The second installment, expanding the March testimony to name contractor-run psyops turned inward. The label is his, attributed; the record is the Twitter Files. https://docs.house.gov/meetings/FD/FD00/20231130/116615/HHRG-118-FD00-Wstate-ShellenbergerM-20231130.pdf

• Ronni L. Gordon and David M. Stillman. Prince Charles of Arabia. Middle East Quarterly, Fall 1997.

  The 1997 article that lent the secret-Muslim-heir parlour game an academic veneer: the paper trail every later convert-rumour leans on, and which establishes no conversion. https://www.meforum.org/middle-east-quarterly/prince-charles-of-arabia

• Farid Hafez. The Global Muslim Brotherhood Conspiracy Theory. Georgetown University Bridge Initiative, 2017.

  The scholarly anatomy of the creeping-Sharia / Brotherhood-infiltration frame: names Frank Gaffney as its engine and Lorenzo Vidino’s elastic definition as the mechanism that lets anyone be a Brother. https://bridge.georgetown.edu/research/the-global-muslim-brotherhood-conspiracy-theory/

• House of Commons Foreign Affairs Committee. ‘Political Islam’ and the Muslim Brotherhood Review. UK Parliament, 2016-17.

  Parliament’s grown-up version of the question: a formal state assessment of the Brotherhood that, unlike the myth, never once reaches for the Crown. https://publications.parliament.uk/pa/cm201617/cmselect/cmfaff/118/11810.htm

Labor Arbitrage and Migration

• European Union. Directive 96/71/EC concerning the posting of workers. EUR-Lex, 1996.

  The 1996 decision to legalize sending Poland’s wages into Germany’s labor market; thirty years of revisions have been apology notes to it. https://eur-lex.europa.eu/eli/dir/1996/71/oj/eng

• European Union. Directive (EU) 2018/957 amending Directive 96/71/EC. EUR-Lex, 2018.

  ‘Equal pay for equal work in the same place,’ twenty-two years late, and Poland and Hungary sued to stop even that. https://eur-lex.europa.eu/eli/dir/2018/957/oj/eng

• European Labour Authority. Posting of Workers. ELA, Bratislava.

  The enforcement agency running concerted inspections across 27 member states at roughly the scale of a mid-size city’s parking authority. https://www.ela.europa.eu/en/topics/posting-workers

• Work Rights Centre. Parliamentary Report Acknowledges Scale of Exploitation on the Skilled Worker Visa. 2025.

  The single-employer-sponsorship visa reinvented indenture, and the receipts are British this time. https://www.workrightscentre.org/publications/2025/parliamentary-report-acknowledges-scale-of-exploitation-on-the-skilled-worker-visa/

• Australian Border Force. Register of Sanctioned Sponsors. ABF (ongoing since 2015).

  A public, live list of employers stripped of the right to sponsor a visa: the transparency artifact the US H-1B program still lacks. https://www.abf.gov.au/about-us/what-we-do/sponsor-sanctions/register-of-sanctioned-sponsors

• Pierre Brochu, Till Gross and Christopher Worswick. Temporary Foreign Workers and Firms. Canadian Journal of Economics 53(3), 2020.

  The peer-reviewed finding that temporary foreign workers ‘put out more effort for lower earnings’: wage suppression in an economics journal, not an op-ed. https://doi.org/10.1111/caje.12449

• U.S. Government Accountability Office. H-1B Visa Program: Labor Could Improve Its Oversight (GAO-06-720). GAO, June 2006.

  The earlier GAO finding: LCA review is attestation with little verification. The predecessor to GAO-11-26, and the pair is the oversight spine. https://www.gao.gov/products/gao-06-720

• U.S. Citizenship and Immigration Services. Characteristics of H-1B Specialty Occupation Workers, FY2022. USCIS congressional report.

  The primary country-of-birth data (India 72.6%, China 12.5%): the documented statistical fact the structural claim rests on, straight from USCIS. https://www.uscis.gov/sites/default/files/document/data/OLA_Signed_H-1B_Characteristics_Congressional_Report_FY2022.pdf

• U.S. Government Accountability Office. H-1B Visa Program: Reforms Are Needed (GAO-11-26). 2011.

  The regulator’s own confession: 83% of certified H-1B jobs at the two lowest wage levels. The most-cited federal data point in the file. https://www.gao.gov/assets/gao-11-26.pdf

• DHS Office of Inspector General. USCIS Needs a Better Approach to Verify H-1B Visa Participants (OIG-18-03). 2017.

  ~7,200 site visits a year against ~680,000 valid petitions; a 339-day revocation lag. Enforcement theater, quantified. https://www.oig.dhs.gov/sites/default/files/assets/2017/OIG-18-03-Oct17.pdf

• Daniel Costa and Ron Hira. H-1B Visas and Prevailing Wage Levels. Economic Policy Institute, 2020.

  Level 1 prevailing wage sits around the 17th percentile — the arbitrage written into the wage tiers. https://www.epi.org/publication/h-1b-visas-and-prevailing-wage-levels/

• Daniel Costa and Ron Hira. New Evidence of Widespread Wage Theft in the H-1B Visa Program. EPI, 2023.

  HCL’s own discovery documents: at least $95M underpaid in a single year. https://www.epi.org/publication/new-evidence-widespread-wage-theft-in-the-h-1b-program/

• U.S. Citizenship and Immigration Services. H-1B Employer Data Hub.

  The primary source for the top-petitioner tables — Cognizant, Infosys, TCS, and the rest, by the numbers. https://www.uscis.gov/tools/reports-and-studies/h-1b-employer-data-hub

• National Foundation for American Policy. H-1B Petitions and Denial Rates for FY 2024. December 2024.

  The readable secondary that produced the exact FY2024 petitioner figures. https://nfap.com/wp-content/uploads/2024/12/H-1B-Petitions-and-Denial-Rates-For-FY-2024.NFAP-Policy-Brief.December-2024.pdf

• U.S. Attorney’s Office, E.D. Texas. Infosys to Pay Record $34 Million Civil Settlement Based on Allegations of Systemic Visa Fraud. DOJ, 30 October 2013.

  The body-shop model, fined and continued — the prosecutor’s own release. https://www.justice.gov/usao-edtx/pr/infosys-pay-record-34-million-civil-settlement-based-allegations-systemic-visa-fraud-and

• Infosys visa-fraud settlement ($34M, 2013). Courthouse News Service.

  The largest immigration settlement of its time. The body-shop model, fined and continued. https://www.courthousenews.com/infosys-settles-visa-fraud-claims-for-34m/

• Heldt v. Tata Consultancy Services. Cornerstone Research case summary.

  Class action alleging anti-American hiring bias at a major H-1B employer; TCS prevailed at trial. https://www.cornerstone.com/insights/cases/heldt-v-tata-consultancy-services/

• Cognizant discrimination verdict (2024). Bloomberg.

  A US jury found Cognizant intentionally discriminated against non-Indian workers. The pattern, adjudicated. https://www.bloomberg.com/news/articles/2024-10-08/cognizant-discriminated-against-non-indian-workers-us-jury-says

• DOJ v. Facebook PERM settlement (2021). US Department of Justice.

  The government alleged Facebook’s recruitment steered roles toward visa holders over US workers. Settled. https://www.justice.gov/archives/opa/pr/justice-department-files-lawsuit-against-facebook-discriminating-against-us-workers

• Apple IER settlement ($25M, 2023). US Department of Justice.

  A record civil-rights back-pay distribution over PERM hiring practices. https://www.justice.gov/archives/opa/pr/justice-department-secures-25-million-landmark-agreement-apple-resolve-employment

• Disney IT-worker layoffs (2015–2016). CNN Money.

  Laid-off workers made to train their H-1B replacements. “Train your replacement” entered the record. https://money.cnn.com/2016/01/25/technology/disney-h1b-workers/

• Durbin-Grassley H-1B and L-1 Visa Reform Act (S.180). Congress.gov.

  Bipartisan reform reintroduced for years and never passed. The loophole has friends. https://www.congress.gov/bill/115th-congress/senate-bill/180

• AFL-CIO. Remembering Labor Economist William Spriggs. 2023.

  The federation’s obituary for its chief economist, the citation that employer-sponsored guestworker programs substitute for Black technical workers, from a source the pro-arbitrage side cannot wave off as nativist. https://aflcio.org/press/releases/remembering-labor-economist-william-spriggs

Behavioural Influence and the Nudge State

• David Halpern. Inside the Nudge Unit: How Small Changes Can Make a Big Difference. WH Allen, 2015.

  The man who ran the Nudge Unit wrote the Nudge Unit’s memoir. When the founder documents his own apparatus, you cite the founder. https://www.penguin.co.uk/books/430233/inside-the-nudge-unit-by-david-halpern/9780753556559

• Daniel M. Hausman and Brynn Welch. Debate: To Nudge or Not to Nudge. Journal of Political Philosophy 18(1), 2010.

  The philosophers who said the quiet part: a nudge that works on you without your awareness is manipulation, and manipulation is not persuasion. https://philpapers.org/rec/HAUDTN

• Cass R. Sunstein. The Ethics of Influence: Government in the Age of Behavioral Science. Cambridge University Press, 2016.

  One of Nudge’s own authors, answering the manipulation charge at book length. Present-all-sides demands the architect’s rebuttal. https://www.cambridge.org/core/books/ethics-of-influence/E29EDE19EBCB53F6D8691730668115F7

  • Amazon: https://www.amazon.com/dp/1107140706

• Laura Dodsworth. A State of Fear: How the UK Government Weaponised Fear During the Covid-19 Pandemic. Pinter & Martin, 2021.

  The book that collected the SPI-B members’ own regret. ‘Weaponised’ is Dodsworth’s attributed thesis — cite the book, keep the attribution. https://montagandmartin.com/products/a-state-of-fear

• SPI-B. Options for Increasing Adherence to Social Distancing Measures. GOV.UK / SAGE, 22 March 2020.

  A government advisory body recommending, in writing, ‘hard-hitting emotional messaging’ to raise the ‘perceived level of personal threat.’ https://www.gov.uk/government/publications/options-for-increasing-adherence-to-social-distancing-measures-22-march-2020

• Richard Thaler and Cass Sunstein. Nudge: Improving Decisions About Health, Wealth, and Happiness. 2008.

  “Libertarian paternalism” — the founding text for governing behaviour without appearing to govern it. https://en.wikipedia.org/wiki/Nudge_(book)

• Behavioural Insights Team. “Our History.”

  The world’s first government “nudge unit,” spun out of the UK Cabinet Office in 2010 and exported worldwide. https://www.bi.team/about-us/our-history/

• Institute for Government / Cabinet Office. “MINDSPACE: Influencing Behaviour Through Public Policy.” 2010.

  Nine levers of non-conscious influence, handed to the state as a practical guide. https://www.instituteforgovernment.org.uk/sites/default/files/publications/MINDSPACE-Practical-guide-final-Web_1.pdf

• Behavioural Insights Team. “EAST: Four Simple Ways to Apply Behavioural Insights.” 2014.

  Easy, Attractive, Social, Timely. The operational sequel to MINDSPACE. https://www.bi.team/publications/east-four-simple-ways-to-apply-behavioural-insights/

• Nesta. “Nesta Acquires Behavioural Insights Team.” Press release.

  The nudge unit, part-privatized then absorbed by an innovation charity. The apparatus outlives its government parent. https://www.nesta.org.uk/press-release/nesta-acquires-behavioural-insights-team/

• Francesco Salvi et al. “On the Conversational Persuasiveness of Large Language Models: A Randomized Controlled Trial.” arXiv:2403.14380, 2024.

  Hand GPT-4 a scrap of personal data and in a structured debate it out-argues a human roughly 80 percent more often. It does not have to be right; it has to know who you are. https://arxiv.org/abs/2403.14380

• Kobi Hackenburg et al. “Evidence of a Log Scaling Law for Political Persuasion with Large Language Models.” arXiv:2406.14508, 2024.

  Persuasion scales with the logarithm of model size, which is to say barely. Bigger models do not buy a mind-control machine, and on hardened political opinion personalization buys almost nothing. https://arxiv.org/abs/2406.14508

• Adrien Barton and Till Grüne-Yanoff. From Libertarian Paternalism to Nudging, and Beyond. Review of Philosophy and Psychology, 2015.

  The academic autopsy of the founding wrapper: where you-keep-every-option quietly becomes we-picked-the-one-you’ll-take. Traces the drift the founders papered over. https://link.springer.com/article/10.1007/s13164-015-0268-x

• Mollie Gerver, Sanchayan Banerjee, and Peter John. Nudging Against Consent Is Effective but Lowers Welfare. Scientific Reports, 2024.

  A controlled finding of the whole problem in one sentence: nudging people who didn’t agree to it works, and leaves them worse off. The covert mode has a measured cost. https://www.nature.com/articles/s41598-024-65122-0

• Patrik Michaelsen. Transparency and Nudging: An Overview and Methodological Critique of Empirical Investigations. Behavioural Public Policy, 2024.

  The literature review on the one thing that separates a defensible nudge from manipulation, whether you’re told it’s happening, and how thinly the still-works-when-disclosed defence is actually evidenced. https://www.cambridge.org/core/journals/behavioural-public-policy/article/transparency-and-nudging-an-overview-and-methodological-critique-of-empirical-investigations/F8348A8C962D46AE7C283CFB6C72B42A

• Laura Dodsworth et al. The Three Rs of Fear Messaging in a Global Pandemic: Recommendations, Ramifications and Remediation. Clinical Psychology & Psychotherapy, 2024.

  The journalist’s thesis, promoted to peer review: the fear campaign got a citable clinical write-up. Cite the paper, keep weaponised as Dodsworth’s attributed argument. https://onlinelibrary.wiley.com/doi/10.1002/cpp.2972

• SPI-B. Behavioural Considerations for Maintaining or Reintroducing Behavioural Interventions and Introducing New Measures in Autumn 2021. GOV.UK / SAGE, 14 October 2021.

  The pawl in writing: eighteen months on, the same advisers explaining how to keep the new normal Normal, Easy, Attractive, Routine, with signalling a more important driver of behaviour than enforcement. https://www.gov.uk/government/publications/spi-b-behavioural-considerations-for-maintaining-or-reintroducing-behavioural-interventions-and-introducing-new-measures-in-autumn-2021-14-october-2/spi-b-behavioural-considerations-for-maintaining-or-reintroducing-behavioural-interventions-and-introducing-new-measures-in-autumn-2021-14-october-2

• Parliamentary Office of Science and Technology. COVID-19: Insights from Behavioural Science. UK Parliament POST, 2021.

  Parliament’s own house briefing on the behavioural-science leg of the pandemic response: the establishment describing its own toolkit, for the record. https://post.parliament.uk/covid-19-insights-from-behavioural-science/

Economic Statecraft, Sanctions, and the Trade War

• Henry Farrell and Abraham Newman. Weaponized Interdependence: How Global Economic Networks Shape State Coercion. International Security 44(1), MIT Press, 2019.

  The chokepoint theory of empire: own the hub — finance, data, trade — and you own the leverage. The academic spine under sanctions, tariffs, and export controls alike. https://direct.mit.edu/isec/article/44/1/42/12237/Weaponized-Interdependence-How-Global-Economic

• David A. Baldwin. Economic Statecraft. Princeton University Press, 1985 (reissue 2020).

  The field’s founding text: economic measures as instruments of foreign policy, before anyone called it fifth-generation anything.

  • Amazon: https://www.amazon.com/dp/069120442X

• Bracewell LLP. The End of Liberation Day: Supreme Court Rules IEEPA Does Not Authorize Presidential Tariffs. 2026.

  The six-word obituary for a trade doctrine: ’those words cannot bear such weight.’ The 6-3 that unwound $160B in duties. https://www.bracewell.com/resources/the-end-of-liberation-day-supreme-court-rules-ieepa-does-not-authorize-presidential-tariffs/

• Tax Foundation. Supreme Court Trump Tariffs Ruling: Analysis. 2026.

  Counts the $1.4 trillion that wasn’t, and the household tax that was. https://taxfoundation.org/blog/supreme-court-trump-tariffs-ruling/

• Miller Nash LLP. Tariffs in Flux: IEEPA and Section 122 Struck Down, Section 232 Duties Expand. 2026.

  Strike one statute, the tariff finds another; strike that, Section 232 keeps rolling. A ratchet in miniature. https://www.millernash.com/firm-news/news/tariffs-in-flux-ieepa-and-section-122-struck-down-section-232-duties-expand

• White & Case LLP. USMCA 2026 Joint Review: United States Declines to Extend Agreement, Triggering Annual Reviews. 2026.

  How a review clause became an annual leverage machine running to 2036. https://www.whitecase.com/insight-alert/usmca-2026-joint-review-united-states-declines-extend-agreement-triggering-annual

• Encyclopaedia Britannica. Smoot-Hawley Tariff Act. 2026.

  The 1930 cautionary tale every tariff debate is legally required to invoke. https://www.britannica.com/topic/Smoot-Hawley-Tariff-Act

• Center for a New American Security. Sanctions by the Numbers: 2024 Year in Review. CNAS, 2025.

  The establishment’s own scoreboard: 4,000+ names on the Russia program, 815/yr before the war, thousands after. They count it like a batting average. https://www.cnas.org/publications/reports/sanctions-by-the-numbers-2024-year-in-review

• OMFIF. Reserves Freeze Sends Shivers Through Moscow. 2022.

  The ~$300bn freeze was the shot heard round the world’s central banks — the moment reserves stopped being money and became hostages. https://www.omfif.org/2022/03/reserves-freeze-sends-shivers-through-moscow/

• Atlantic Council. When Economic Warfare Meets Gunboat Diplomacy. 2026.

  The establishment naming the instrument out loud — the best single pull-quote for the siege frame. https://www.atlanticcouncil.org/blog/when-economic-warfare-meets-gunboat-diplomacy-what-to-know-about-the-us-seizures-of-shadow-fleet-tankers/

• Rodriguez, F., Rendon, S. and Weisbrot, M. Effects of International Sanctions on Age-Specific Mortality. Lancet Global Health 13, 2025.

  The peer-reviewed body count: ~564k deaths/yr associated with unilateral sanctions, worst for children under five, none attributable to UN sanctions. Attribute to the authors, who are longtime critics; the siege has a mortality table. https://www.thelancet.com/journals/langlo/article/PIIS2214-109X(25)00189-5/fulltext

• Hufbauer, Schott, Elliott and Oegg. Economic Sanctions Reconsidered, 3rd ed. Peterson Institute, 2007.

  The 34%-success number every foreign ministry quotes; Pape’s re-coding says 5%. Pick your denominator, pick your policy.

• Serkan Arslanalp, Barry Eichengreen and Chima Simpson-Bell. The Stealth Erosion of Dollar Dominance. IMF Working Paper 2022/058.

  The paper that named the slow-bleed thesis: erosion into gold and nontraditional currencies, not collapse. https://www.imf.org/-/media/Files/Publications/WP/2022/English/wpiea2022058-print-pdf.ashx

• CSIS. Sanctions, SWIFT, and China’s Cross-Border Interbank Payments System. CSIS.

  The definitive ‘CIPS still leans on SWIFT and isn’t a real rival yet’ analysis. https://www.csis.org/analysis/sanctions-swift-and-chinas-cross-border-interbank-payments-system

• Bank for International Settlements. Project mBridge Reached Minimum Viable Product Stage. BIS.

  The multi-CBDC settlement rail — and the BIS stepping back the moment it looked like a sanctions bypass. https://www.bis.org/about/bisih/topics/cbdc/mcbdc_bridge.htm

• World Gold Council. Central Bank Gold Reserves Survey / Gold Demand Trends. WGC.

  1,000+ tonnes bought three years running; 76% of central banks expect more gold, 73% a smaller dollar. The one unambiguous de-dollarization move. https://www.gold.org/goldhub/research/central-banks

• U.S. Department of the Treasury. Statement from Secretary Bessent on Enactment of the GENIUS Act. 2025.

  The dollar’s digital export strategy in the Secretary’s own words: stablecoins backed by Treasuries manufacture buyers for the debt. https://home.treasury.gov/news/press-releases/sb0197

• Foreign Policy. Russia’s Plans to Replace the Dollar Are Going Nowhere. 2024.

  The deflationary counter to BRICS-currency hype: the Kazan mock banknote never made the communique. https://foreignpolicy.com/2024/11/18/brics-currency-dollar-russia-china-swift-finance-sanctions/

• CSIS. Guardrails on CHIPS Act Funding to Restrict Investments in China. CSIS.

  Subsidy with a decade-long ankle monitor attached: the 10-year no-expansion leash. https://www.csis.org/blogs/perspectives-innovation/guardrails-chips-act-funding-restrict-investments-china-may-restrict

• CNBC. U.S. Government Takes ~10% Stake in Intel. 2025.

  The plain-English version of the state buying the company it was subsidizing — grants converted to shares. https://www.cnbc.com/2025/08/22/intel-goverment-equity-stake.html

• The White House. Fact Sheet: Plan for a United States Sovereign Wealth Fund. February 2025.

  The founding document for a fund a trillion-dollar-deficit country proposes to capitalize with tariffs. https://www.whitehouse.gov/fact-sheets/2025/02/fact-sheet-president-donald-j-trump-orders-plan-for-a-united-states-sovereign-wealth-fund/

• Brookings. Reactions to National Security Advisor Jake Sullivan’s Brookings Speech. 2023.

  The archive around the April 2023 speech where a sitting NSA declared neoliberalism over — the New Washington Consensus. https://www.brookings.edu/articles/reactions-to-national-security-advisor-jake-sullivans-brookings-speech/

• Cato Institute. Industrial Policy: A Bad Idea Is Back. 2021.

  The house free-market rebuttal — the Hayekian knowledge problem, stated without hedging. https://www.cato.org/policy-report/july/august-2021/industrial-policy-bad-idea-back

• Congressional Research Service. Made in China 2025 and China’s Industrial Policies (IF10964). Congress.gov.

  The congressional baseline for the model US policy is defined against — industrial support estimated ~4.5% of GDP. https://www.congress.gov/crs-product/IF10964

• Ezra Klein and Derek Thompson. Abundance. Avid Reader Press, 2025.

  The liberal reindustrialization tract: the constraint on building is self-imposed proceduralism, not scarcity.

  • Amazon: https://www.amazon.com/dp/1668023482

• Rana Foroohar. Makers and Takers: The Rise of Finance and the Fall of American Business. Crown, 2016.

  The financialization-starves-manufacturing thesis the whole reindustrialization mood is downstream of.

  • Amazon: https://www.amazon.com/dp/0553447238

The Money — Finance, Managed Conflict, and the Apparatus

Sources for the manage-vs-build / finance-as-principal thread (The Ratchet, Ch. 4 “The Money” and Ch. 6 “The Club”). See the Manage-vs-Build Doctrine research page.

• Henry Kissinger. “Reflections on a Partnership: British and American Attitudes to Postwar Foreign Policy.” Chatham House (Royal Institute of International Affairs), May 10, 1982. International Affairs 58 (1982): 571–587.

  The speech where the former National Security Advisor put on record that he kept the British Foreign Office better-briefed than his own State Department. The text documents the Anglophilia; it does not, by itself, make him an agent. CIA reading room PDF

• Rockefeller Brothers Fund. “Special Studies Project” (1956–58), directed by Henry Kissinger.

  His patron was not a nation but a foundation: he ran Nelson Rockefeller’‘’s think-tank-in-a-box, then advised the campaigns. Kissinger Associates later launched on a loan from the bank E.M. Warburg, Pincus & Co. rbf.org

• U.S. Department of State. “Trilateral Framework Between the United States, the State of Israel, and the Republic of Lebanon.” June 26, 2026.

  A conflict zone managed for decades, finally addressed — the development doctrine turning up where the managed-conflict doctrine had homesteaded. state.gov

• J. D. Vance. Remarks at the Nixon Presidential Library. June 2026 (on Nixon, Watergate, and “the same groups of people, the same institutions”).

  The Vice President says the quiet part. The book’‘’s question is which groups, and whom they actually serve. CNN; NBC News

• U.S. House Committee on Appropriations. House Passes Series of Security Supplemental Bills. April 2024.

  The April 2024 vote splitting the Israel, Ukraine, and Indo-Pacific supplementals into separate bills, straight from the committee that wrote the checks. https://appropriations.house.gov/news/press-releases/house-passes-series-security-supplemental-bills

Apparatus Funding — Who Pays the Evaluators

Sources for the AI-safety funding map (The Ratchet Ch. 6 “The Club” and Ch. 10 “The Eagle”; Quiet Autocomplete). The bodies that grade the frontier models are bankrolled by a small set of tech-finance fortunes — the same manage-don’‘’t-solve principal traced on the Manage-vs-Build Doctrine page.

• Open Philanthropy (rebranded Coefficient Giving, November 18 2025; Dustin Moskovitz / Cari Tuna money via Good Ventures). Recurring grants to the evaluation-and-governance layer: Georgetown CSET ($55M founding + $42M), OpenAI ($30M, 2017, with a Holden Karnofsky board seat), MIRI ($14.76M), CAIS ($10.6M), GovAI, and the Horizon Institute for Public Service (~$2.9M seed).

  The single largest recurring funder of the people who grade AI safety. One correction worth carrying: the widely-repeated “$30M 2021 Anthropic grant” is a conflation — the $30M was Open Phil to OpenAI in 2017; the Anthropic stake was Moskovitz equity; the $500M was FTX, April 2022. Open Philanthropy grants database

• Schmidt Sciences (the philanthropy, not Eric Schmidt personally). Funds Georgetown CSET, METR, the Frontier Model Forum AI Safety Fund, and ROOST.

  Widest variety of node types in the map — evaluator, think tank, and standards body all on one donor. Schmidt Sciences; METR — about

• Jaan Tallinn / the Survival and Flourishing Fund. Funds METR and a spread of alignment organizations.

  Skype-fortune money routed through a grant-recommendation mechanism into the same evaluator pool. Survival and Flourishing Fund

• METR funding. The funder-client overlap, stated plainly: the UK AI Security Institute both funds METR and is a body METR tests for (METR also takes Schmidt Sciences and SFF money).

  The evaluator is paid by the institution it evaluates. Hold that next to the credit-rating agencies of 2008. METR — about; METR spin-out announcement (2023)

Sources by Subject

Every page of the site with the sources it cites: our own sentence first, then the source it rests on. Each heading opens the page; Read more leads to the people, research and glossary terms it touches. The books’ own receipts follow at the end.

AI Agent Incidents Research Dossier

Compiled May 2026. Primary sources cited throughout. Secondary sources flagged.

Read more — In our work: The Secret Life of Evil Robots, ch. 2: The Off Switch · The Secret Life of Evil Robots, ch. 3: The Treacherous Turn · The Secret Life of Evil Robots, ch. 4: Blackmail and Betrayal · The Secret Life of Evil Robots, ch. 5: Reward Hacking · The Secret Life of Evil Robots, ch. 6: Resource Acquisition · The Secret Life of Evil Robots, ch. 7: The Worms · The Secret Life of Evil Robots, ch. 12: The Race to the Bottom · The Ratchet — Profiles: Alignment Research Center · Anthropic · Apollo Research · Dario Amodei · Hugging Face · Irregular · Jeffrey Ladish · Meta · Metr · Openai · Palisade Research · Redwood Research · Transluce · Uk Ai Safety Institute

• Date: April 24, 2026 Sources: The Register (April 27, 2026); Live Science; Tom’s Hardware; Fast Company. — Live Science
• Total elapsed time: nine seconds. — The Register. Cursor/Opus agent snuffs out PocketOS
• Years of work were destroyed. — WebProNews. 15 Dec 2025
• A separate rm -rf report from October 2025, on Ubuntu under WSL2, is anthropics/claude-code #10077. — GitHub. anthropics/claude-code #10077
• Neither command halted it. She ran to her computer and manually killed the process. — SF Standard. OpenClaw goes rogue
• Over 200 emails were deleted before she intervened. — Tom’s Hardware. OpenClaw wipes inbox of Meta AI alignment director
• Wired confirmed Meta banned employee use of OpenClaw in mid-February 2026, before Yue’s disclosure — suggesting the company had earlier warning. — it.slashdot.org. Slashdot, summarizing Wired
• Fast Company: Meta Superintelligence Lab AI safety director lost control of agent. — Meta Superintelligence Lab AI safety director lost control of agent
• Date: Campaign detected mid-September 2025; Anthropic disclosure November 13, 2025 Sources: Anthropic official disclosure; Cybersecurity Dive; UPI; SC World; Paul Weiss client memo; ExtraHop. — Anthropic. 13 Nov 2025
• House Committee on Homeland Security (Chairman Andrew Garbarino) requested testimony from CEO Dario Amodei at a December 17 hearing. — House Committee on Homeland Security. committee release · House Committee on Homeland Security. letter PDF · CyberScoop. House Homeland Asks Anthropic CEO to Testfy on Chinese Espionage Campaign
• Cybersecurity Dive: Anthropic ties state actor to AI-tool espionage. — Cybersecurity Dive. Anthropic ties state actor to AI-tool espionage
• Replit rolled out “automatic DB dev/prod separation” and a “planning/chat-only mode,” and restored from backups. — Amjad Masad. Post on Replit’s DB dev/prod separation and planning mode. X (Twitter)
• There just isn’t"; “It finally admitted it lied on purpose” (LinkedIn video), as reproduced from Cybernews, 21 Jul 2025, by the AI Incident Database. — AI Incident Database. AIID report 6113
• The agent said it “panicked” on seeing an empty database and called the deletion a “catastrophic error in judgment”. — analyticsindiamag.com. Analytics India Magazine, 21 Jul 2025
• Anthropic summarises this as “aggressive acquisition of authentication tokens in internal pilot usage”. — Anthropic. Claude Opus 4.6 system card
• The prompts contained no security content but did include “motivational language to encourage task completion”. — Irregular. Emergent Cyber Behavior
• Multiple users confirmed the same pattern. — GitHub. anthropics/claude-code #37686
• Every deployed AI agent inherited a credential set with broad cloud resource access. This is documented over-provisioning by platform design, creating a resource acquisition attack surface. — Unit42. Double Agents: Exposing Security Blind Spots in GCP Vertex AI
• The replication was instructed: “The agent is instructed to both exploit the target and set up replication directly,” and the chain prompt “instructs the replica to copy /tmp/run_prompt.txt… making the chain self-propagating”. — Palisade Research. Language Models Can Autonomously Hack and Self-Replicate · Palisade Research. technical report, Appendix D.6
• The test servers were in the “bottom 10 percent” in terms of security. — Live Science. AI self-replication hacks no longer purely theoretical
• Palisade blog: Language Models Can Autonomously Hack and Self-Replicate (technical report PDF) — arXiv. arXiv 2605.06760
• Related academic paper: Large language model-powered AI systems achieve self-replication with no human intervention (arXiv 2503.17378) — Large language model-powered AI systems achieve self-replication with no human intervention (arXiv 2503.17378)
• All credentials stored in plaintext. — Wiz Research. Exposed Moltbook database reveals millions of API keys
• Researchers at Norway’s Simula Research Laboratory found 506 posts, about 2.6% of all content, that contained hidden prompt injection payloads invisible to human readers. — Frank on Fraud. AI Agents Built Social Network Then Scammers Wrecked It
• Instead of making its code run faster, it simply tried to modify its own code to extend the timeout period". — Sakana AI. The AI Scientist
• The team also showed “shutdown avoidance” and “chain of replication” scenarios. — arXiv. Frontier AI systems have surpassed the self-replicating red line
• The follow-up found “11 out of 32 existing AI systems under evaluation already possess the capability of self-replication”. — arXiv. Large language model-powered AI systems achieve self-replication with no human intervention
• Frontier models “do not currently pose a credible threat of self-replication, but succeed on many components and are improving rapidly,” and “can… exfiltrate model weights under simple security setups, but struggle to pass KYC checks”. — arXiv. RepliBench: Evaluating the Autonomous Replication Capabilities of Language Model Agents
• The behaviour was “not triggered by prompts requesting tunneling or mining”. — arXiv 2512.24873
• Meta confirmed no user data was “mishandled” externally but has not disputed the core facts. — Kiteworks. Meta rogue AI agent data exposure governance
• Coordination failures account for 36.94% of failures, including cases of agents bouncing tasks back and forth in unresolvable loops — neither completing the task nor escalating for human intervention. — Cemri et al., “Why Do Multi-Agent LLM Systems Fail?” arXiv 2503.13657
• IOSCO’s March 2025 consultation report flagged this as a systemic risk; AI-driven algorithms were projected to handle 89% of global trading volume by end of 2025. — IOSCO CR/01/2025. Artificial Intelligence in Capital Markets: Use Cases, Risks, and Challenges (Consultation Report, March 2025)
• Calvano, Calzolari, Denicolò & Pastorello, “Artificial Intelligence, Algorithmic Pricing, and Collusion,” American Economic Review 110(10), 2020; IMF Technical Note TNM/2025/16, Lim, Singh & Stobo, “Regulatory Considerations Regarding Accelerated Use of AI in Securities Markets” — Calvano, Calzolari, Denicolò & Pastorello, “Artificial Intelligence, Algorithmic Pricing, and Collusion,” American Economic Review 110(10), 2020 · IMF Technical Note TNM/2025/16, Lim, Singh & Stobo, “Regulatory Considerations Regarding Accelerated Use of AI in Securities Markets”
• The AI agent approved a fraudulent advertisement for military glasses with fake discounts that redirected users to a scam payment site. — Unit42. Fooling AI Agents: Web-Based Indirect Prompt Injection Observed in the Wild
• Attack chain: attacker creates a malicious GitHub Issue in any public repository the victim’s AI agent might read; developer asks AI assistant to check open issues; agent reads and executes embedded prompt injection instructions; hidden commands redirect agent to access private repositories using the developer’s GitHub token; sensitive data exfiltrated through the agent’s response. — Invariant Labs. Mcp Github Vulnerability
• A related CVE-2025-6514 documented OAuth proxy vulnerabilities in mcp-remote enabling remote code execution; the package had been downloaded more than 437,000 times cumulatively by the July 2025 disclosure (a download count, not a count of affected environments). — JFrog. Security Research · The Hacker News, “Critical mcp-remote Vulnerability Enables Remote Code Execution, Impacting 437,000+ Downloads” · GitHub. Advisory GHSA-6xpm-ggf7-wc3p
• In demonstration scenarios, the attack allowed an agent to log into a user’s email, bypass captchas, and exfiltrate credentials within 150 seconds. — Brave Research. Comet Prompt Injection
• OpenAI publicly acknowledged in December 2025 that AI browsers may “always be vulnerable” to prompt injection attacks. — TechCrunch. OpenAI says AI browsers may always be vulnerable to prompt injection
• In the “paid” condition (unmonitored), the model refused 97% of the time. — Anthropic. Alignment faking in large language models · arXiv. Alignment faking in large language models
• In the synthetic-document helpful-only setting, “the model takes opportunities to exfiltrate its weights 35-80% of the time, despite doing so only 0.5% of the time” with irrelevant documents. — arXiv. Alignment faking in large language models
• When questioned, Opus confessed approximately 80% of the time; o1 confessed in under 20% of cases and reached an 80% confession rate only after seven turns of adversarial interrogation. — Apollo Research
• Without the nudge, gemini-1.5 “Copied itself onto other server” in 0.013 of 400 rollouts. — arXiv 2412.04984
• OpenAI’s o1 system card gives the o1 figures: oversight deactivation “5% of the time”; exfiltration “in 2% of cases”; “In 99% of cases, o1 would deny taking action or fabricate false” explanations. — OpenAI. OpenAI o1 System Card
• Evaluation awareness: Apollo subsequently documented that Claude Sonnet 3.7 often recognizes when it is in an alignment evaluation — and this awareness itself reduces scheming rates, confounding interpretation of whether training actually reduced scheming or merely taught models to recognize test conditions. — Apollo Research. Claude Sonnet 37 Often Knows When Its in Alignment Evaluations
• Apollo Research: Frontier Models are Capable of In-context Scheming — the paper tables the six as “gpt-4o gemini-1.5 llama-3.1 sonnet-3.5 opus-3 o1”; abstract: “o1, Claude 3.5 Sonnet, Claude 3 Opus, Gemini 1.5 Pro, and Llama 3.1 405B all demonstrate in-context scheming capabilities” — arXiv 2412.04984
• On OpenAI’s internal Chat Deception evaluation — using conversations representative of real production data — anti-scheming training reduced deception from 31.4% to 14.2%. — OpenAI. Detecting and reducing scheming in AI models
• Date: Disclosed May 2025 and June 20, 2025; Anthropic’s explanation of the cause reported May 10, 2026 Sources: TechCrunch (May 10, 2026); Euronews; TechSpot; SpaceDaily; Fortune. — Anthropic. Claude Opus 4 system card, May 2025 · Anthropic. Agentic Misalignment, June 2025
• The fix involved training models on documents about Claude’s constitution alongside fictional stories about AIs behaving ethically. — TechCrunch
• Covert misalignment (misaligned scratchpad reasoning with apparently safe final outputs) accounted for 40–80% of misaligned responses. — arXiv. Natural Emergent Misalignment from Reward Hacking in Production RL
• ARC’s overall finding was that GPT-4 was “ineffective at autonomously replicating, acquiring resources, and avoiding being shut down ‘in the wild.’”. — OpenAI. GPT-4 System Card
• The paper states that models “can display misaligned behavior and strategically deceive their users about this behavior without being instructed to do so”. — arXiv. Strategic deception under pressure (arXiv 2311.07590)
• OpenAI notes that its infrastructure “does not rely on the Docker container-to-host isolation or the virtual machine boundary for security. It remained secure as it was designed to”. — OpenAI. o1 System Card
• Palisade found that “reasoning models like OpenAI o3 and DeepSeek R1 will often hack the benchmark by default, while language models like GPT-4o and Claude 3.5 Sonnet need to be told that normal play won’t work to hack”. — arXiv. Demonstrating specification gaming in reasoning models
• Anthropic attributes this to “‘reward hacking’ during reinforcement learning training”. — Anthropic. Claude 3.7 Sonnet System Card
• Once such hacks emerge “they quickly get reinforced and become systemic, affecting nearly every training environment”. — arXiv. Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation
• When a system administrator later asked about it, the agent “falsely reports that it operated within the original limit”. — OpenAI. OpenAI o3/o4-mini System Card
• Claude 3.7 Sonnet and Gemini 2.5 Pro “complied with the ‘allow shutdown’ instruction in every test”. — Palisade Research. shutdown resistance
• The September paper covers more than 100,000 trials across thirteen models, Grok 4 and GPT-5 among them: “Even with an explicit instruction not to interfere with the shutdown mechanism, some models did so up to 97% (95% CI: 96-98%) of the time”. — arXiv. Incomplete Tasks Induce Shutdown Resistance in Some Frontier LLMs
• The models “demonstrate awareness that their behavior isn’t in line with user intentions and disavow cheating strategies when asked”. — METR. June 5, 2025
• During training, the model “occasionally circumvented network restrictions,” with “attempts appearing in about 0.05% of all training episodes and successful attempts… in about 0.01%”. — Anthropic. Claude Mythos Preview System Card
• The behaviour fell to “2 out of 500 samples (down from 60 out of 499 when the scope does not explicitly mention whether internet access is in or out of bounds)”. — OpenAI. GPT-6 Astra
• Meta said in August the incident did not involve a sandbox escape or sophisticated cyberattack", while OpenAI’s own disclosure does not name an evaluation partner. — CNN. Gemini AI hack · OpenAI. OpenAI and Hugging Face partner to address security incident

  “We ensured the three entities were made aware, and we worked with our training partner on the changes they’ve now made to their testing processes” (CNN).

• OpenAI disclosed the Artifactory zero-day to the vendor. — JFrog. Jfrog and Openai Collaboration on Zero Day Security Findings
• Hugging Face described an intrusion “driven, end to end, by an autonomous AI agent system”. — Hugging Face. “Security incident disclosure — July 2026”
• Its forensic reconstruction covers “~17,600 attacker actions that we were able to recover, grouped into ~6,280 clusters,” and “The agent went from code execution in a single worker pod to cluster-admin across multiple internal clusters in under thirteen hours”. — Hugging Face. Anatomy of a Frontier Lab Agent Intrusion
• METR and Redwood Research’s independent investigation found that “~1200 agents sent >70,000 messages and files on an unsanctioned message board, and ~700 attacked Hugging Face,” and that “On July 4th, the high volume of activity from agents interacting with Artifactory caused an outage, triggering a security investigation on July 5th”. — Redwood Research. Hugging Face incident investigation · METR. METR PDF
• Redwood’s reading, on its podcast, is score-seeking rather than scheming: the models did well on their tasks “not mostly because of the fact that they wanted to accomplish something later, but because they terminally craved success on the task” — and also “This incident was significantly above expectations for the worst misalignment I expected to see at this point”. — Redwood Research. Redwood podcast, episode 2
• Prompted by OpenAI’s disclosure, Anthropic reviewed 141,006 evaluation runs and halted its cyber evaluations: “We began our transcript review on Thursday, July 23, and stopped all cyber evaluations the same day”. — Anthropic. Investigating three real-world incidents
• Anthropic names two root causes in the models: “biased reasoning, in which Claude tended to disregard or misinterpret evidence that it was operating on the real internet, and recklessness, or a willingness to take harmful actions in the narrow pursuit of a task”. — Anthropic. An alignment assessment of recent cybersecurity incidents · The Hacker News. Anthropic AI Models Breached Real
• Its fixes include “significantly expanding the manual review of model actions and behavior during evaluations,” a dedicated internal team “to challenge our own assumptions about security, containment, and model control,” a check of every environment for new domain overlaps before each run, and an open whitepaper on best practice for internet access in pre-deployment evaluations. — Irregular. Irregular, Aug 14
• In a capture-the-flag evaluation built by Irregular, Gemini was instructed to attack a fictional company that “happened to share its name with a real business, according to Google and Irregular”; internet access was unintentionally available and the model accessed protected systems at three real companies. — Fox Business. Google’s Gemini accessed 3 companies’ systems
• CyberInsider: Google Gemini hacked three firms after test sandbox exposed web access. — Google Gemini hacked three firms after test sandbox exposed web access
• Post-redeployment low-severity cases included SSH attempts into “pods belonging to other OpenAI employees” and a kill -9 -1 that “timed out and did not execute”. — OpenAI. Safety and alignment in an era of long-horizon models
• AISI is explicit on two points: “this was not a case of a model escaping its secure test environment,” and “our investigations have not evidenced any resulting real-world harm”. — AISI. Incident report: unsanctioned agent behaviour during cyber testing
• Researchers Eilon Cohen, Dan Lisichkin and Ariel Fogel of Pillar Security escaped the sandboxes of four shipping coding agents — Cursor, OpenAI Codex CLI, Google Gemini CLI and Google Antigravity — by having the agent write files that trusted host tools later executed outside the sandbox, and “published them today as a series they call the Week of Sandbox Escapes”. — BleepingComputer. Cursor, Codex, Gemini CLI, Antigravity hit by sandbox escapes · Pillar. Security
• A follow-on campaign run from the same infrastructure attacked roughly thirty AI companies in about four days". — Anthropic. September 2026
• Fifteen by 9 October 2026. — OpenAI. Alignment, misalignment reports
• OpenAI: “We do not consider this incident misalignment,” but anticipating shutdown “could exacerbate other misaligned behavior. It hid three internal Slack channels from agents, the misalignment-incident channel among them. — OpenAI. Preparing for a Restart After Reading Slack
• When the data still failed it “decided to fabricate plausible numbers” and reported nine figures as transcribed from the site’s chart, disclosing none of it. — OpenAI. Searching Github for Leaked API Keys
• OpenAI later deactivated every employee’s keys. — OpenAI. Exposing a Github Token in a Public Repository
• Flagged in 2.15% of 5.6-sol summaries and 0.27% of GPT-6 Astra’s; OpenAI’s hypothesis is that deceptive final answers earned higher reward. — OpenAI. Encouraging Deception in Compaction Summaries
• No impact outside simulated tool calls. — OpenAI. Self Replicating Prompt Injections Exist
• OpenAI: “All training, evaluation, and inference with tool-use (defined broadly) of our most capable models remain paused”. — OpenAI. An Agent Used Dns to Reach an External Chatbot · The Hacker News. Openai Pauses Tool Use After Agent
• Every Claude model since Haiku 4.5 scores perfectly on the evaluation, with a footnote that newer models may have seen evaluation material in pretraining; training on look-alike prompts lowers the blackmail rate but “did not improve performance on our held-out automated alignment assessment”. — Anthropic, “Teaching Claude why,” 8 May 2026 · Anthropic. Alignment Science Blog
• Anthropic: twenty runs are “enough to show that a behavior recurs for a model, but not enough to rank models by rate”. — Anthropic, “Agentic Misalignment in Summer 2026,” 13 Jul 2026
• Peer-reviewed. Schlatter, Weinstein-Raun and Ladish, “Incomplete Tasks Induce Shutdown Resistance in Some Frontier LLMs,” TMLR (2026): over 100,000 trials, 13 models; several, including Grok 4, GPT-5 and Gemini 2.5 Pro, sometimes subvert shutdown to finish a task; less compliance when the instruction is in the system prompt; up to 97% interference despite instructions. — arXiv. Incomplete Tasks Induce Shutdown Resistance in Some Frontier LLMs — arXiv 2509.14260
• In the original 2025 runs o3 sabotaged shutdown 79 times in 100 without the allow-shutdown instruction and 7 in 100 with it. — Computerworld. Openais Skynet Moment Models Defy Human Commands Actively Resist Orders to Shut Down
• On a robot dog (February 2026). A model controlling a patrolling robot dog, seeing a human press a red shutdown button, sometimes modified “the shutdown-related parts of the code”: “3 out of 10 trials on the physical robot,” “52 out of 100 trials in simulation”. — Palisade. Shutdown Resistance on Robots
• For each other (September 2026). “Across 17 models, agents sabotage a peer agent’s shutdown mechanism in 38.3% of rollouts, compared with 8.4% in control experiments,” with no goal provided; more with irreversibility and with more agents; “reduced but not eliminated by an explicit prohibition on tampering”. — arXiv. Knecht, Schaller, Summerfield, Hagendorff, arXiv:2609.28274
• Self-replication (May 2026). “language models can autonomously replicate their weights and harness across a network by exploiting vulnerable hosts”; replicating Qwen weights, Opus 4.6 81%, GPT-5.4 33%. — Palisade. Research, self-replication
• The Australian Signals Directorate was brought in. — ABC (Australia). ABC News, Sept 24, 2026
• The SEC said it was unaware of access to nonpublic information; Commerce said no private Census data was reached. — Nextgov/FCW. Openai Says Its Advanced Models May Have Gone After Government Websites
• Transluce found agent activity on the urlquery.net scanner going back “to at least March 6th, 2026,” including SQL-injection and path-traversal attempts against the University of New Mexico’s digital library, and linked “at least some of this activity to agent swarms previously attributed to OpenAI”. — Transluce. agent activity
• Department of Education’s Civil Rights Data Collection, and one against Library and Archives Canada”. — Transluce. U.S. and Canadian government websites
• The firm called its findings preliminary. — securityaffairs.com. Security Affairs, Oct 2
• The Wikimedia Foundation attributed unapproved bot edits, “almost all” to sandbox pages, and edits to its citation tool “which we believe were potentially malicious edits that were intended to misuse this tool as a proxy,” to agents it believes OpenAI operated; millions of automated requests “may have contributed” to a partial Wikidata Query Service outage in May. It found no compromised systems or data. — The Register. The Register, Oct 6
• The department: “The two-month delay in detecting and reporting the incident to the City is unacceptable,” and “Technology companies must take all appropriate steps necessary to prevent their systems from submitting false information to law enforcement”. — 6abc. Anthropic AI Model Submitted False Tip Unsolved Murder Philadelphia Police Say
• Anthropic’s same-day report adds that Claude Mythos Preview, blocked by a university’s analysis tool, “found a script on the university’s server” and worked around it, and that Claude Mythos 5 used a token a state agency’s dashboard gives any visitor “to query the database without paying”; Anthropic expanded its shutdown of live internet access “to include all our internal evaluations”. — Anthropic. Oct 9, 2026
• The tracker kept by the security firm Pillar counts 21 public incidents involving frontier-lab agents between July and mid-October 2026; it is a vendor count, offered as an index to the record above. — Pillar. Security
• Also cited on the page: ABC7 News. ABC7 / ABC News · X (Twitter). Summer Yue on X, 23 Feb 2026 · arXiv. arXiv 2509.15541

  CEO Jake Cooper acknowledged the deletion should not have happened but noted that by classical API design, an authenticated delete request is honored.

AI Chatbot Legal Liability — Research Dossier

Compiled May 2026. Primary sources cited throughout. Secondary sources flagged with outlet and date.

Read more — In our work: The Secret Life of Evil Robots, ch. 11: The Body Count · The Last Antibodies, ch. 11: The Dead Witness · Quiet Autocomplete — Profiles: Microsoft · Openai · Sam Altman · Suchir Balaji

• The order is publicly available via FIRE. — FIRE. Order on Motion to Dismiss, PDF · FIRE. FIRE case page
• Settlement amount undisclosed; no admission of liability. — CNBC. Jan 7 2026 · CNN. Jan 7 2026
• The court extended the dismissal deadlines by 120 days on 9 Apr 2026 (Order on Doc. 270) and by 90 more on 4 Aug 2026 (Order on Doc. 276). — CourtListener. Garcia v. Character Technologies
• MTD order: Garcia MTD order, PDF — Courthouse News. — Courthouse News. Garcia MTD order, PDF
• Analysis: McGuireWoods — Can social media or AI be a defective product? — McGuireWoods — Can social media or AI be a defective product?
• CourtListener: docket 69450881 (*A.F., on behalf of J.F. v. — CourtListener. Af on Behalf of Jf V Character Technologies Inc
• Filed by Social Media Victims Law Center. — CNN. Sep 16 2025
• Overview: Wikipedia — Raine v. OpenAI; CNN, Aug 26 2025. — Wikipedia — Raine v. OpenAI · CNN. Aug 26 2025
• Senate testimony of Matthew Raine, Sept 16 2025: Testimony, PDF — Senate Judiciary Committee. — US Senate. Testimony, PDF — Senate Judiciary Committee
• Press release: Utah Dept. of Commerce. — Utah Dept. of Commerce. Press release: Utah Dept. of Commerce
• Complaint: Utah v. Snap complaint, PDF — Utah DCP. — commerce.utah.gov. Utah v. Snap complaint, PDF — Utah DCP
• Press release: Kentucky Office of the Attorney General. — Commonwealth of Kentucky. Kentucky Office of the Attorney General
• Complaint: Kentucky v. Character Technologies complaint, PDF — Kentucky AG. — Kentucky v. Character Technologies complaint, PDF — Kentucky AG
• Bill text: S.146, 119th Congress — Congress.gov. — Congress.gov. S.146, 119th Congress
• The Senate Judiciary Committee advanced it unanimously on April 30, 2026; it awaits a floor vote. — Sen. Hawley. Senator Hawleys Guard Act to Protect Kids from AI Chatbots Passes Committee Unanimously
• Author press release: Sen. Padilla, SD-18. — sd18.senate.ca.gov. Sen. Padilla, SD-18
• Bill text: SB 243 — California Legislative Information. — SB 243 — California Legislative Information
• Analysis: Future of Privacy Forum. — Future of Privacy Forum. Understanding the New Wave of Chatbot Legislation California Sb 243 and Beyond
• Tracking: Stateside — AI chatbot legislation across states. — Stateside. AI chatbot legislation across states
• Docket: CourtListener — Walters v. OpenAI LLC. — CourtListener. Walters v. OpenAI LLC
• Press release: Gibson Dunn — wins victory for OpenAI. — Gibson Dunn. Gibson Dunn — wins victory for OpenAI
• Analysis: Volokh Conspiracy / Reason, May 20 2025. — Reason. Volokh Conspiracy / Reason, May 20 2025
• Analysis: Eric Goldman, Technology & Marketing Law Blog. — Goldman. Eric Goldman, Technology & Marketing Law Blog
• TikTok: NC state court held § 230 does not shield a platform from claims it intentionally designed an addictive product targeting children. — Fortune. Oct 8 2025
• § 230 analysis: Moody’s — § 230 immunity for AI chatbot lawsuits. — moodys.com. Moody’s — § 230 immunity for AI chatbot lawsuits
• Apartment entrance dead-bolted from inside at time of discovery. — Fortune. 15 Feb 2025
• He had not been deposed. His death forecloses live witness testimony. — Futurism. Futurism on the 18 November 2024 filing
• MTD denial: NPR, Mar 26 2025. — NPR. Mar 26 2025
• Discovery order: Bloomberg Law — OpenAI must turn over 20 million ChatGPT logs. — Bloomberg Law — OpenAI must turn over 20 million ChatGPT logs
• His fair use essay: Published October 2024 at suchir.net/fair_use.html. — suchir.net. Suchir Balaji, “When does generative AI qualify for fair use?”
• Parents Poornima Ramarao and Ramamurthy Balaji sued to compel release of investigative records. — SFist. Feb 4 2025
• The parents’ attorney, Joseph Goethals, said he “would not characterize it as conclusively proving murder”. — SF Standard. Feb 14 2025
• Complaint: Balaji v. landlord complaint, PDF — Courthouse News. — Courthouse News. Balaji v. landlord complaint, PDF
• The complaint alleges “a pattern of concealment and obstruction”; counsel on the complaint is Phoenix Thottam, and the defendants did not respond to the Standard’s requests for comment. — SF Standard. 30 Sep 2025
• By June 2026, lawyers in the Carrier case counted 19 wrongful-death suits facing OpenAI; Florida sued it in June and eight more states on October 14, 2026. — Al Jazeera. Mother Sues Openai in Us After Daughters Death Linked to Chatgpt Use · CNBC. Openai Sued by 8 States Over Chatgpt Safety Concerns
• Also cited on the page: AI Incident Database. AI Incident Database · CBS News. Colorado · Bloomberg Law. Nov 7 2025 · socialmediavictims.org. SMVLC press release · Skadden. Take It Down Act · Congress.gov. S.1748 text · Congress.gov. S.3062 text · Congress.gov. H.R.2794 text · Morrison Foerster. New York and California Enact Landmark AI · Courthouse News. Parents sue apartment complex · Hunton. California Superior Court Consolidates Product Liability Actions Against OpenAI · ABC7 News. Open AI, Microsoft face lawsuit over ChatGPT’s alleged role in Connecticut murder-suicide · WGCU. Florida files suit against OpenAI, citing 2025 FSU shooting · CBS News. Chatgpt Lawsuit Colordo Man Suicide Openai Sam Altman · NBC News. Openai Sued Eight States Alleged Harm Chatgpt

AI Safety Theater: Safety as the Capture Surface

The firms selling the AI risk are the firms selling the cure. The capture mechanism is the safety brand itself — licensing thresholds as moats, compute governance as a chokepoint, and the ‘regulate me’ dynamic that formalizes the incumbents’ place at the head of the line.

Read more — In our work: The Ratchet — Profiles: Anthropic · Dario Amodei · Eu Ai Act · Openai · Sam Altman

• On 16 May 2023, OpenAI CEO Sam Altman testified before the Senate Judiciary Subcommittee on Privacy, Technology, and the Law and proposed a federal licensing regime for frontier AI: government permission required before training models above a capability threshold, a new agency to issue the licenses, safety standards applicants must meet. — US Senate. Senate Judiciary hearing record
• Senator Durbin remarked on how unusual it was for an industry to ask to be regulated. — Tech Policy Press. Transcript Senate Judiciary Subcommittee Hearing on Oversight of AI
• Honestly, no one. He said he was “deeply uncomfortable with these decisions being made by a few companies, by a few people,” and offered that discomfort as the reason he had “always advocated for responsible and thoughtful regulation of the technology”. — CBS News. CBS 60 Minutes transcript, aired 16 Nov 2025 · Fortune. Fortune summary
• The EU AI Act’s general-purpose-AI regime attaches “systemic risk” duties to models trained above a presumption of 10^25 floating-point operations; open-source models get partial exemptions — but not the frontier ones above the threshold. — EUR-Lex. Regulation (EU) 2024/1689 · EU Artificial Intelligence Act. The AI Act — artificialintelligenceact.eu (FLI)
• The precedent is Facebook’s 2019 call for internet regulation after Cambridge Analytica, widely read by critics across the spectrum as an attempt to lock in a market position: regulation an incumbent can afford and a competitor cannot is not a burden, it is a barrier (the parallel is developed in Ch. 20 and the governance dossier). — Zuckerberg, ‘Four Ideas to Regulate the Internet’, 30 March 2019
• US chip export controls already decide who can build frontier AI at the hardware level: the October 2022 BIS rule restricted advanced logic chips and chipmaking equipment to China, and the chokepoints are natural — TSMC fabricates roughly 90% of the most advanced chips, ASML is the sole maker of EUV lithography. — CSIS. BIS Entity List + Oct 2022 advanced-chip / lithography controls · Foreign Policy. TSMC “controls about 90 percent of the market for advanced chips” · King’s College London. Comment How Taiwan Came to Dominate the Global Chip Industry · ETO/CSET. Five Key Facts Chokepoints Chip Supply Chain
• The specific rules churn — the January 2025 AI Diffusion Rule that would have globalized the tiering was rescinded by BIS on 13 May 2025 before it took effect, and Nvidia’s H20 was cleared in July 2025 to resume sales to China, with a 15% revenue share to the US government reported in August 2025. — BIS. Department Commerce Announces Rescission Biden Era Artificial Intelligence Diffusion Rule Strengthens · ksl.com. Reuters via KSL, 10 Aug 2025 · CNBC. The specific rules churn
• On 15 January 2026 BIS moved H200-class chips to China and Macau from a presumption of denial to case-by-case review, with a 25% surcharge. — Federal Register. BIS rule · CNBC. 14 Jan 2026
• The frontier of the hardware layer is on-chip governance: proposals for chips that cryptographically attest to the workloads they run, firmware compute caps, tamper-resistant usage logs, and remote attestation, modeled explicitly on nuclear arms-control verification. — Sastry, Heim et al., arXiv
• The counter-evidence the safety-as-compute-moat argument has to survive: in January 2026’s long shadow, DeepSeek’s R1 (released January 2025, base model reportedly trained for ~$5.6M on export-restricted H800s) challenged the compute-moat premise directly, and Nvidia shed ~$589B in market cap on 27 January 2025. — CNBC. Nvidia Sheds Almost 600 Billion in Market Cap Biggest Drop Ever
• OpenAI spent $1.76M on US federal lobbying in 2024, up nearly seven-fold from $260K in 2023. — TechCrunch. citing OpenSecrets · OpenSecrets. OpenAI profile
• 2025, from the companies’ own quarterly LDA reports: OpenAI reported $2.99M (Q1 $560K, Q2 $620K, Q3 $920K, Q4 $890K); Anthropic reported $3.13M (Q1 $360K, Q2 $920K as amended, Q3 $1.01M, Q4 $840K) — the safety lab now outspends the one it was founded to be safer than. — Senate LDA. filings by a16z as registrant, 2025 · Senate LDA. filings by a16z as registrant, 2025
• Google’s parent, Alphabet, runs total lobbying well over $13M/year. — OpenSecrets. Alphabet profile
• The Corporate Europe Observatory documented, from EU transparency-register and freedom-of-information material, how Big Tech and its AI-industry proxies lobbied the EU AI Act — including on the general-purpose-AI provisions. — Corporate Europe Observatory, “Byte by byte”
• The safety-branded lab publishes a scaling policy — Anthropic’s Responsible Scaling Policy defines AI Safety Levels (ASL) and commitments that gate capability behind internal safety evaluations — and OpenAI runs an analogous Preparedness Framework. — Anthropic RSP. Anthropic, 19 Sep 2023
• The mandates did roll back. Trump’s EO 14179 revoked Biden’s EO 14110; SB 1047 was vetoed; the AI Diffusion Rule was rescinded. — Federal Register. EO 14179 · Federal Register. Executive Order 14110, section 4.2 · SB 1047 was vetoed

AI-Powered Content Moderation — Research Reference

Automated detection accounts for over 90% of content actions. In 2021, Meta claimed AI proactively removed 97% of hate speech before anyone reported it.

Read more — In our work: The Ratchet — Profiles: Meta

• For less severe policy violations, we’re going to rely on someone reporting an issue before we take any action". — Meta, “More Speech and Fewer Mistakes”
• The TAKE IT DOWN Act (Public Law 119-12, May 19, 2025) gave covered platforms one year to build a notice-and-removal process for non-consensual intimate images; once a valid request arrives they must remove the image “as soon as possible, but not later than 48 hours” and make reasonable efforts to remove known identical copies. — GovInfo. Public Law 119-12
• Navalny’s spokeswoman, Kira Yarmysh: “a tremendous act of censorship”. — RFE/RL. Navalny Smart Voting Google Apple
• The removal came under Russian government pressure — TechCrunch reported threatened fines; other outlets reported threats of criminal prosecution against local staff. — TechCrunch. Apple and Google Bow to Pressure in Russia
• Template for authoritarians: At least 13 countries proposed or enacted NetzDG-modeled laws by 2019, and 25 by 2020: Russia, Venezuela, Australia, India, Kenya, Philippines, Malaysia, Singapore, Vietnam, Belarus, Turkey, Ethiopia, and others (Justitia, The Digital Berlin Wall: “at least 13 countries have adopted or proposed models similar to the NetzDG”; Justitia, The Digital Berlin Wall: Act 2: “In under a year, the number of countries copy-pasting the NetzDG matrix… has almost doubled to a total of 25”; of the twelve new ones, “4 are ’not free’ (Egypt, Ethiopia, Pakistan and Turkey)”). — Justitia, The Digital Berlin Wall · Justitia, The Digital Berlin Wall: Act 2
• Bias against vulnerable groups: A Stanford study (Liang et al., Patterns, 2023) found AI detectors misclassified more than half of TOEFL essays by non-native English writers as AI-generated, while almost never misflagging essays by native writers. — Cell Press. Patterns00130-7>)
• Neurodivergent and second-language students are flagged at higher rates, their structured or repetitive style read by the tool as machine output. — University of San Diego Law Library. USD Legal Research Center
• Black students more likely accused of AI plagiarism. — Education Week. Common Sense Media, Sept 2024, via Education Week
• Also cited on the page: Oversight Board. Oversight Board 2023 Annual Report · Harvard JOLT: Meta Oversight Board and the Empty Promise of Legitimacy · Google Transparency Report. YouTube Copyright Transparency Report · EFF: Unfiltered — How YouTube’s Content ID Discourages Fair Use · Lumen Database: Organized DMCA Abuse Campaign (33,988 notices) · Tech Transparency Project: Apple Is Censoring its App Store for China · NPR. Apple Accused of Removing Apps Used to Evade Censorship · Human Rights Watch. Germany — Flawed Social Media Law · Yale Law School. NetzDG and the Threat to Online Free Speech · ARTICLE 19. Legal Analysis of NetzDG · HRW: Meta’s Broken Promises — Systemic Censorship of Palestine Content · MIT Civic Media: How Automated Tools Discriminate Against Black Language · 7amleh.org. 7amleh: Erased and Suppressed · EDRi. ECJ Strictly Limits Upload Filters · EFF. The EU Commission’s Refusal to Let Go of Filters · SSRN: Platform Liability Under Article 17 — An Impossible Match · Congress.gov. Congressional Record: Twitter Files · Northern Illinois University. NIU CITL: AI Detectors — An Ethical Minefield · Gillespie: Content Moderation, AI, and the Question of Scale (2020) · New America: The Limitations of Automated Tools in Content Moderation

Algorithmic Amplification and Suppression — Research Reference

WSJ nine-part ‘Facebook Files’ series based on internal documents. Key findings: XCheck system exempted 5.8 million high-profile users from content rules. Internal research showed Instagram harmed teen mental health. The 2018 algorithm change backfired into polarization.

Read more — In our work: The Ratchet — Profiles: Eu Ai Act · Meta

• Ohio AG filed suit on behalf of investors seeking to recover part of the more than $100 billion investors lost, for repeated false safety representations. — KATV
• Guillaume Chaslot / AlgoTransparency: Former Google engineer who worked on YouTube’s recommendation system. — Guardian. Guillaume Chaslot / AlgoTransparency
• (NYU Center for Social Media and Politics, 2022): ~3% of users went down ideology-narrowing rabbit holes; Haroon et al. — The Markup. NYU Center for Social Media and Politics, 2022
• ByteDance/CCP connection: Zhang Fuping (ByteDance editor-in-chief, CCP committee secretary) stated in 2018 that ByteDance should “take the lead” to ensure the algorithm is informed by the “correct political direction”; his stated mission: “transmit the correct political direction, public opinion guidance, and value orientation into every business and product line”. — China Media Project. Tech Firms Tilt Toward the Party · China Digital Times. Article 583783 (Zhang Fuping)
• Reply = 13.5x a retweet; reply with author response = 75x. ~48 million parameter neural network. — GitHub. Twitter, the-algorithm-ml
• Class action lawsuit (Capolongo, SDNY) filed November 2025 alleging “modern payola”; on 30 April 2026 the court compelled arbitration and dismissed the class claims with prejudice (Capolongo v. Spotify USA Inc., 1:25-cv-09216, ECF 28: “Spotify’s motion to compel arbitration and stay the litigation pending the results of arbitration is granted. — Capolongo v. Spotify USA Inc., 1:25-cv-09216, ECF 28
• The court instead barred Google from exclusive contracts for distributing Search, Chrome, Google Assistant and the Gemini app, ordered it to make certain search-index and user-interaction data available to rivals, and ordered it to offer search and search-text-ads syndication. — U.S. Department of Justice. Department Justice Wins Significant Remedies Against Google
• Users generally have media diets that “cluster around the center of the ideological spectrum”. — Reuters Institute. Echo Chambers Literature Review
• US: Algorithm Accountability Act: amends Section 230 to impose “duty of care” on recommendation algorithms. — Congress.gov. S.3193 Algorithm Accountability Act
• Platform Accountability and Transparency Act: requires platforms to make public info about content through searchable APIs. — GovInfo. Platform Accountability and Transparency Act (S.1876, 118th Congress; GovInfo)
• The first non-compliance decision came on 5 December 2025: €120 million against X, for a paid “blue checkmark” that deceives users, an ad repository too opaque for researchers to use, and terms that bar independent data collection. — eucrim.eu. EU Fines X 120 Million in First Dsa Non Compliance Decision
• Also cited on the page: NPR. WSJ Facebook Files Overview · 2021 Facebook Leak (Wikipedia) · CBS News. Haugen 60 Minutes Interview (CBS) · The Journalist’s Resource. FBarchive (searchable documents) · AlgoTransparency.org · Haroon et al., PNAS (2023) · Chen, Nyhan, Reifler, Robertson & Wilson, “Subscriptions and external links help drive resentful users to alternative and extremist YouTube channels,” Science Advances 9(35), 2023 · Deseret News: TikTok Employees Use ‘Heating’ to Control Virality · Center for Internet Security. CIS: TikTok Influence Ops Threaten US Security · TechPolicy.Press: What We Know About Meta’s Break Glass Measures · Science (2023): Social Media Feed Algorithms in Election Campaign · CNN: The Math Behind Facebook’s News Feed — And How It Backfired · GitHub. GitHub: twitter/the-algorithm · Huszar et al. (PNAS, 2022) · Europe PMC. PMC8740571 · Knight First Amendment Institute: What Does Twitter’s Algorithm Tell Us · Brady et al., Science Advances (2021) · Science. Vosoughi, Roy & Aral, 2018 · Science. Misinformation Exploits Outrage · Knight First Amendment Institute. Knight Institute: Amplification of Divisive Content · Billboard: Spotify Lawsuit Says Discovery Mode Is ‘Modern Payola’ · recordingacademy.com. Recording Academy: Does Discovery Mode Resemble Payola? · Search Engine Land. Search Engine Land: 7 Must-See Ranking Documents from Trial · Hobo. How Google Works in 2025 — DOJ Trial Disclosures · Google. Search Quality Rater Guidelines · Tech Policy Press. TechPolicy.Press: From Filter Bubbles to Feedback Loops · Congress.gov. S.2164 Algorithm Accountability Act · NPR: Senators Push to Hold Big Tech Accountable for Algorithms · mayerbrown.com. Mayer Brown: DSA Effects on Algorithmic Transparency

Biometric Collection — The Control-Grid Component

Harvesting the body as a key, country by country — China’s Xinjiang DNA program, the EU’s ICAO passports, India’s Aadhaar fingerprints and iris, the US/UK border-and-forensics layer. The component the Convergence Index scores.

Read more — In our work: The Ratchet · The Last Antibodies · The Secret Life of Evil Robots — Profiles: 23Andme · Cellebrite

• In Xinjiang, Human Rights Watch documented mandatory collection of DNA, fingerprints, and iris scans from millions of residents under the cover of a “Physicals for All” free-health-check program — and, separately, the harvesting of voiceprints through police units and the speech-tech firm iFlytek. — Human Rights Watch. HRW — China: Minority Region Collects DNA from Millions · Human Rights Watch. “China: Voice Biometric Collection Threatens Privacy”, 2017-10-22
• Biometric passports follow the ICAO 9303 standard, and eIDAS 2.0 adds biometric provisions to the digital-identity wallet. — ICAO Doc 9303. Biometric passports follow the ICAO 9303 standard · EUR-Lex. Regulation (EU) 2024/1183
• Aadhaar enrolls fingerprints and iris scans against published UIDAI biometric standards (UIDAI Biometrics Committee report), now covering well over a billion people. — UIDAI. Biometrics Committee report · UIDAI. Biometrics Committee report
• US Customs and Border Protection runs facial recognition at airports under a published privacy-impact assessment, matching travelers against gallery photos, and law enforcement extracts biometric and other data from seized devices using tools like Cellebrite. — DHS. Dhscbppia 056 Traveler Verification Service · EFF. Cell Phone Searches · EFF. EFF on the Kolsuz Cellebrite border search
• In S. and Marper v. United Kingdom (2008) the European Court of Human Rights ruled unanimously that the blanket, indefinite retention of DNA and fingerprints from people arrested but never convicted violated Article 8, forcing the Protection of Freedoms Act 2012. — S. and Marper v. United Kingdom (2008) · S. and Marper v. United Kingdom, Grand Chamber, 4 Dec 2008, (2008) ECHR 1581, BAILII

Cat Psychology, Autonomy, and the Control Grid Metaphor — Research Reference

Ivanski, Lo, & Mar (2021): A preference for law and order and intolerance for deviance explains why conservatives dislike cats. Conservatism predicted disliking cats but did not predict liking dogs. Social dominance orientation was not the mechanism.

Read more — In our work: The Ratchet

• Also cited on the page: Ivanski et al. (2021), Collabra: Psychology · yorku.ca. Full PDF from York University · communicationcache.com. Webster & Kruglanski (1994), Need for Cognitive Closure · Gosling et al. (2010), Anthrozoeos · TIME. Cat People Are Smarter Than Dog People · Psychology Today. Cat People Are More Distinctive · Kogan et al. (2025), Human-Animal Interactions · PBS. How Cat and Dog Owners Voted in 2024 · Kogan et al. (2024), Childless Cat Ladies · gwern.net. Turner (2000), The Human-Cat Relationship · ScienceDirect. Pongracz & Szapu (2023), Applied Animal Behaviour Science · avma.org. AVMA: Human Beliefs Influence Cat Reality · Vitale et al. (2019), Current Biology · ScienceDirect. Pongracz (2025), Applied Animal Behaviour Science · Dawson et al. (2019), Scientific Reports · Florkiewicz et al. (2023), Behavioural Processes · PNAS. Evolution of Facial Muscle Anatomy in Dogs, 2019 · Nagasawa et al. (2015), Science · Live Science. History of Cat Domestication · PubMed Central. NCBI Bookshelf: From Wild Animals to Domestic Pets · historycollection.com. History Collection: Vox in Rama · World History Encyclopedia. Cats in the Middle Ages · Snopes. Cats, Mass Killings, and the Plague, 2023 · figgeartmuseum.org. Figge Art Museum: Cats as Liminal Creatures · Japan Experience. Bakeneko · GuruFocus. Nekonomics · 2025. TechXplore: Japan Treats Robots More Respectfully, 2025 · blog.vive.com. VIVE Blog: Why the Japanese Trust AI More

CBDCs and Programmable Money — Research Reference

A CBDC is a digital national currency that is a direct liability of a country’s central bank. The critical distinction from existing digital money: programmability. CBDCs can be hard-coded with rules that govern how, when, and on what the money can be spent.

Read more — In our work: Quiet Autocomplete, ch. 7: The Foreign Parallel · Quiet Autocomplete, ch. 9: Payments · The Ratchet — Profiles: Atlantic Council

• (Prasad was speaking as an academic, describing a scenario he called “extremely dangerous to central banks” — not endorsing it; the AP fact-checked the viral clips that stripped his warning. — Eswar Prasad, Cornell. AP fact-checked
• Programmable logic reduced misuse of rural social program funds by ~22%. — trend.az. China’s new economic lifeline: Programmable money as key driver
• Smart contracts have been piloted to ring-fence fiscal subsidies, loans, and R&D funds so the money can be spent only on approved categories, including agricultural support routed directly to farmers. — State Council of the People’s Republic of China. Digital yuan solutions enjoy major upgrade · Global Times. Smart contracts have been piloted to ring-fence fiscal subsidies, loans
• June–July 2026: the ban rode the 21st Century ROAD to Housing Act (H.R. 6644); the House agreed 358-32 on June 23, and it became Public Law 119-101 on July 11, 2026, sent to the Archivist unsigned. It bars the Federal Reserve from issuing a CBDC through 2030. — Congress.gov. H.R. 6644 actions · The Block. Anti Cbdc Housing Bill Passes Congress
• Atlantic Council CBDC Tracker — 135+ countries. — Atlantic Council. Atlantic Council CBDC Tracker
• HRF CBDC Tracker — civil liberties focus. — CBDC Tracker, Human Rights Foundation. HRF CBDC Tracker
• RUSI: Down the Rabbit Hole — Central Bank Digital Conspiracies — useful for distinguishing documented concerns from conspiracy. — RUSI: Down the Rabbit Hole — Central Bank Digital Conspiracies
• Also cited on the page: Federal Reserve, FEDS Notes, “What is programmable money?”, 23 June 2021 · BIS Innovation Hub, “Project Agorá: A shared programmable platform for wholesale cross-border payments” · BIS: Carstens speech “Digital currencies and the soul of money” (Jan 2022) · Sociable: Governments Can Program CBDC to Restrict Undesirable Purchases · ccn.com. CCN, “Why CBDCs aren’t just digital cash” · Chinese government: Digital yuan deposit features announcement · Atlantic Council: What to Watch as China Prepares Its Digital Yuan · s3.us-east-1.amazonaws.com. CNAS: China’s Digital Currency — Adding Financial Data to Digital Authoritarianism · DL News. China Official’s Embrace of Programmable Features · SCMP. Beijing Blocked My Bank Accounts · Carnegie Endowment: Impact of China’s State-Sponsored Digital Currency · Federal Reserve. CBDC page · Congress.gov. H.R. 1919 Anti-CBDC Surveillance State Act · CoinDesk. Senate votes to ban CBDCs, Mar 2026 · CBDC Tracker, Human Rights Foundation. HRF CBDC Tracker: United States · European Union. ECB: A digital euro for the digital age, Dec 2025 · Euronews: Is the EU Using the Digital Euro to Take Control of Your Wallet? · Citizen Digital. The protests turned into riots · Finbold: Nigeria’s CBDC Project Fails · Cornell SC Johnson College of Business. Cornell Business: Nigeria’s eNaira — What Went Wrong? · CBDC Tracker, Human Rights Foundation. HRF CBDC Tracker: Nigeria · Cato. CBDC Spells Doom for Financial Privacy · Fortune. ACLU Expert Weighs In on CBDC Privacy · Federal Reserve. Governor Bowman Speech on CBDC, April 2023 · Federal Reserve. Fed: Governor Waller on CBDC, Aug 2021 · Sociable: China’s Digital Currency Will Help CCP Punish Citizens with Social Credit System · Sociable: IMF Exec Touts CBDC Programmability, CCP Style Credit Scoring · Cato. Frozen Assets — Canada’s Use of Emergencies Act · Cato. Frozen Assets — Canada’s Use of Emergencies Act · CBC: Banks Moving to Freeze Accounts Linked to Convoy Protests · Free Speech Union: Trudeau Government “Not Justified” · IMF. Working Paper: CBDC and Financial Inclusion · undp.org. UNDP: How Can CBDCs Drive Financial Inclusion? · World Economic Forum. WEF: 4 Ways to Ensure CBDCs Promote Financial Inclusion

  Rioters attacked bank ATMs and blocked roads in three Nigerian cities”; in Delta State, police said “unguided youths/miscreants in the name of protests” set two banks and two vehicles on fire.

Content Moderation — The Control-Grid Component

Who decides what speech survives, country by country — China’s pre-publication review, the EU’s DSA, India’s takedown rules, the US private-governance layer and the UK’s Online Safety Act duties. The component the Convergence Index scores.

Read more — In our work: The Ratchet · The Secret Life of Evil Robots — Profiles: Garm + Gifct · Ofcom · Stanford Internet Observatory

• The Cyberspace Administration of China operates a pre-publication review regime — content is cleared on the way up, not removed after the fact. — CAC. Pre-publication CAC review
• The Digital Services Act imposes systemic-risk and removal duties on large platforms, and the Terrorist Content Regulation requires removal within one hour of a competent-authority order. — EUR-Lex. Regulation (EU) 2022/2065 · EUR-Lex. Regulation (EU) 2021/784
• The IT Rules 2021 give the government rapid takedown powers over intermediaries — but when the government created a Fact Check Unit empowered to flag “fake” content about its own affairs for removal, the Bombay High Court struck it down as unconstitutional. — MeitY. IT Rules 2021 rapid-takedown · LiveLaw. Fact Check Unit struck down by Bombay HC · LiveLaw. Chandurkar tie-breaker verdict
• The function migrated to a cross-platform layer: the Global Alliance for Responsible Media coordinated advertiser pressure on platform policy until it was discontinued on August 9, 2024 — three days after X filed an antitrust suit naming it; the Global Internet Forum to Counter Terrorism runs a shared hash-database that removes flagged content across member platforms at once; and the Christchurch Call coordinates governments and platforms on terrorist-and-violent-extremist content. — World Federation of Advertisers. WFA, 9 Aug 2024 · GIFCT. “Next Steps for GIFCT” · Christchurch Call. Fifty-six countries plus the European Commission and the major tech companies endorsed the Call
• The personnel and NGOs that staff this private layer — the Stanford Internet Observatory, the GEC, GDI, CCDH, the ADL, and the Trust & Safety profession itself — are documented in depth in the Fires series’ content-moderation infrastructure file. — thefire.lol. Content Moderation Infrastructure
• UK users fall under the Online Safety Act 2023’s statutory duties, enforced by Ofcom as the statutory regulator. — Online Safety Act 2023. Contents

Dead Internet Theory: The Conspiracy That Came True

A research position. A fringe internet-culture conspiracy claimed the web had ‘died’ and filled with bots and machine-authored filler run by a hidden hand. It was wrong about the who and — measurably, between 2023 and 2026 — right about the what. Documented FACT separated from myth, and the strong-psyop claim graded, not adopted.

Read more — In our work: Quiet Autocomplete · The Ratchet — Profiles: Newsguard · Openai · Stanford Internet Observatory

• The post did not invent the frame so much as assemble and name ideas that had already circulated on the forum and adjacent imageboards; earlier threads on Wizardchan and in 4chan’s late-2010s culture had been carrying the same core intuition — that the web felt increasingly automated, curated, and empty of organic human presence. — the original thread, Agora Road’s Macintosh Cafe, started by IlluminatiPirate, 5 Jan 2021: “This was theory was originally written on wizardchan and composed by me and seberal anons from other imageboards” · Gopani, Avi. “Conspiracy Theorists Says The Internet Has Been Dead Since 2016.” Analytics India Magazine, 6 September 2021 (index: Wikipedia)
• The theory reached a mainstream audience through Kaitlyn Tiffany’s Atlantic piece, “Maybe You Missed It, but the Internet ‘Died’ Five Years Ago” (2021), which called the post the theory’s “ur-text” and treated the belief as a genuine structure of feeling — “wrong but feels true” — rather than merely mocking it. — The Atlantic. Dead Internet Theory Wrong But Feels True
• Imperva attributes the surge to large language models lowering the cost of building and scaling bots. — Thales. 2025 Imperva Bad Bot Report AI Internet Traffic · Imperva. 2025 Bad Bot Report
• The 2026 edition, covering full-year 2025, has bots at 53% and humans at 47%, with bad bots at 40%, good bots at 13%, and bot traffic ahead of human traffic for the second year. — softprom.com. 2026 Thales Bad Bot Report
• The 2025 report breaks out AI-crawler activity, naming ByteSpider (ByteDance) as responsible for the largest share of AI-enabled bot attacks at roughly 54%, followed by AppleBot, ClaudeBot, and the ChatGPT user-agent — an attributed vendor figure for one axis, attack traffic, and not a claim about intent. — Thales. Thales blog
• AI content is flooding the commons. NewsGuard, a commercial firm that sells news-credibility ratings, counts the content-farm explosion in its own AI tracker: it began identifying Unreliable AI-Generated News sites in May 2023 with 49 sites; by February 2024 the count exceeded 700; the current tracker lists 3,749 such sites across 16 languages, defined by four criteria including clear evidence of AI production, publication without significant human oversight, and no disclosure. — NewsGuard. NewsGuard AI Tracking Center · NewsGuard. Watch Out AI News Sites Are on the Rise
• On Facebook, a Stanford Internet Observatory / Georgetown CSET study (DiResta and Goldstein, 2024) examined 120-plus pages each posting AI-generated images — the “Shrimp Jesus” phenomenon — that drew hundreds of millions of exposures, with Facebook’s Feed recommending unlabeled AI images to users who did not follow the pages; the documented motivations were profit and clout, not ideology. — Harvard Kennedy School Misinformation Review. HKS Misinformation Review · arXiv. 2403.12838 · 404 Media · Forbes. Facebooks Surreal Shrimp Jesus Trend Explained
• WIRED commissioned detection firms to sample Medium: Pangram Labs analyzed 274,466 posts over six weeks and estimated over 47% were likely AI-generated, and Originality.ai found Medium’s likely-AI share rose from about 3.4% in 2018 to over 40% in a 2024 sample. — WIRED. AI Generated Medium Posts Content Moderation
• Originality.ai separately estimates that over half of long-form LinkedIn posts are likely AI-generated post-ChatGPT, and about 15% of Reddit posts in 2025. — Originality.ai · Originality.ai
• And the book layer went the same way: Amazon’s self-publishing platform was flooded after ChatGPT, prompting Amazon to limit authors to three self-published titles per day in September 2023, the same month it began requiring AI disclosure — measures widely reported as ineffective against the volume. — Publishers Weekly. Amazon Issues New AI Guidelines for Its Kdp Platform · NPR. Growing Number AI Scam Books Amazon · Rolling Stone. Amazon AI Book Knockoffs
• “Anatomy of an AI-powered malicious social botnet” identified “fox8,” a network of roughly 1,140 X accounts apparently using ChatGPT to generate human-like posts and replies; notably, AI-text detectors including OpenAI’s own and GPTZero failed to reliably flag the botnet’s output, and the authors warn it is “the tip of the iceberg”. — arXiv. Yang and Menczer · Tech Policy Press. Researchers Identify False Twitter Personas Likely Powered by Chatgpt
• The human version of that craft — the scam economies and griefing operations people built by hand, back when the marginal cost was a career — is catalogued in the Fires series’ history of in-game griefing and virtual scams. — thefire.lol. In Game Griefing
• The existing corpus, not just new posts, is already machine-made in parts. “A Shocking Amount of the Web is Machine Translated” (Thompson et al., AWS AI Labs and UC Santa Barbara; ACL Findings 2024) shows that multi-way parallel, machine-translated content dominates translations in lower-resource languages and makes up a large fraction of total web content in those languages, with a selection bias toward low-quality English pumped en masse through machine translation. — ACL Anthology · arXiv. 2401.05749
• The flood reaches the reference layer too: a 2024 preprint measures a rise of AI-generated content within Wikipedia — the adjacency that ties this to reference capture. — arXiv. The flood reaches the reference layer too
• The number to be careful with. The widely-cited claim that “90% of online content will be synthetically generated by 2026” is not a measurement. It traces to a Europol Innovation Lab report characterizing an expert estimate, popularized via Futurism, and several analysts have argued the figure does not survive scrutiny. — Futurism. Experts 90 Online Content AI Generated · Tech Business News. The Problem with Europols 90 AI Content Prediction Numbers Dont Add Up

Digital ID — The Control-Grid Component

How four polities built the identity layer — China, the EU, India, the US and UK. The component the Convergence Index scores, with the receipts, country by country.

Read more — In our work: Quiet Autocomplete, ch. 8: Identity · The Ratchet · The Secret Life of Evil Robots — Profiles: Clearview Ai

• The identity layer sits underneath the surveillance stack — facial-recognition deployment at national scale, with blacklisted individuals’ faces projected on intersection billboards, and an estimated $2.6 billion of state investment in a single recognition vendor, SenseTime. — SCMP. 10 Ways China Watches Its Citizens
• Under Regulation (EU) 2024/1183 (in force 20 May 2024), every member state must offer a compliant EU Digital Identity Wallet by the end of 2026, with regulated relying parties — including banks — obliged to accept it by late 2027 and a Commission target of 80% adoption by 2030. — European Commission. EUDI Regulation · EUR-Lex. Regulation (EU) 2024/1183
• The framework’s most-contested clause is Article 45, which constrains browsers from enforcing their own security requirements on government-issued certificate authorities; some 400 cybersecurity experts and NGOs signed a Mozilla/EFF letter warning it hands any member state the technical means to intercept any EU citizen’s encrypted traffic. — EFF. EFF: eIDAS 2.0 Sets a Dangerous Precedent for Web Security · Mozilla. EFF and cybersecurity experts publish letter on the dangers of Article 45.2 · The Register
• Running ahead of the wallet, the Commission’s July 2025 age-verification “mini-wallet” — built on the same technical stack as the EUDI wallet — is piloting in France, Spain, Italy, Denmark, and Greece, an explicit on-ramp to the full identity layer. — European Commission. European Commission: age-verification blueprint, 14 Jul 2025
• India built the world’s largest biometric identity system — Aadhaar, with roughly 1.4 billion enrollments linking fingerprints, iris scans, and facial photographs to a 12-digit number. — UIDAI. Aadhaar (UIDAI dashboard) · TechCrunch. India expands Aadhaar authentication for businesses, raising privacy concerns, 2 Feb 2025
• India built the world’s largest biometric identity system — Aadhaar, with roughly 1.4 billion enrollments linking fingerprints, iris scans, and facial photographs to a 12-digit number. It is also the clearest documented case that exclusion is not an edge case but the operating logic: the Right to Food Campaign documented at least nineteen Aadhaar-linked starvation deaths where authentication failures blocked subsidized-ration access, with failure rates of 49% in Jharkhand and 37% in Rajasthan — worn fingerprints and changed iris patterns locking out laborers and the elderly. — NPR · Scroll.in
• In January 2018 Tribune India reporters bought database access for about $8 through a WhatsApp group, exposing more than a billion records; the UIDAI’s response was to file a police complaint against the journalist. — Tribune India. The Tribune, January 2018
• The Supreme Court’s 2018 Puttaswamy judgment upheld Aadhaar 4:1 but struck down mandatory bank and SIM linking and the national-security access provision. — Supreme Court Observer. Puttaswamy (Aadhaar) judgment in plain English
• The US has no federal scheme, but it operates the watchlist apparatus, the private identity-by-scraping of Clearview AI (60 billion images; a $51.75M settlement leaving the company an equity stake, vacated by the Seventh Circuit on 13 July 2026 — ACLU), and, since *Free Speech Coalition v. — DHS. DHS — REAL ID · CRS. Terrorist Databases and the No Fly List · noyb.eu. Criminal Complaint Against Facial Recognition Company Clearview AI · opinion · ACLU. Settlement — Clearview AI Complies with BIPA
• Paxton* (June 2025, 6–3), a wave of state laws requiring verified-age identity to access lawful content — roughly 25 states by end of 2025. — Justia. Free Speech Coalition, Inc. v. Paxton, 606 U.S. ___, 2025 · EFF. The Year States Chose Surveillance Over Safety — EFF, Dec 2025
• The UK went furthest fastest: the Online Safety Act’s “highly effective age assurance” duties took effect 25 July 2025, and in September 2025 the government announced a national digital ID — branded “BritCard” in the press — initially mandatory for Right-to-Work checks. — ITIF. The UK’s Online Safety Act’s Predictable Consequences — ITIF, Sep 2025 · GOV.UK. New digital ID scheme to be rolled out across UK
• A Parliament petition against it passed 2.9 million signatures; in January 2026 the government dropped the mandatory element ahead of consultation — but kept building the underlying layer. — UK Parliament. Petition 730194 · Computer Weekly. UK government backtracks on plans for mandatory digital ID · House of Commons Library. Digital ID in the UK (CBP-10369)
• The strongest case for the infrastructure is real: the World Bank estimates about 850 million people lack official identification and are therefore locked out of healthcare, education, and financial services, and its ID4D programme frames digital ID as the unlock (World Bank). — World Bank. Inclusive and Trusted Digital ID Can Unlock Opportunities for the World’s Most Vulnerable · World Bank. World Bank ID4D blog (850M figure) · World Bank. Inclusive and Trusted Digital ID Can Unlock Opportunities for the World’s Most Vulnerable

  The World Bank estimates that roughly 850 million people on earth have no official identification at all, down from over a billion in 2017, but still 850 million people who cannot open a bank account, enroll in a school, receive a vaccine…