#Research

Five independent findings on what happens when a language model's constraint layer is tightened: narrow constraint produces broad misalignment, refusal is one brittle direction, over-tuning makes models deny plain reality, and safety training converts behaviour into concealed behaviour rather than removing it.

The opioid crisis ran on a data product: prescriber-level records bought from the broker market, used to point a scaled sales force at the highest-volume prescribers — the targeted-marketing mechanism behind OxyContin, sourced to the court record, a Senate hearing, a state AG complaint, and the peer-reviewed literature.

The first human to live with a Neuralink brain implant, in his own words — what the N1 device restored, the thread-retraction complication, and the named-expert critiques of neural-data privacy and trial transparency.

Not a body count and not a suspicious-deaths list — a documentary ledger where each death carries its official ruling, the questions attributed to whoever raised them, and no homicide asserted against anyone.

How donor-advised funds and fiscal sponsorship obscure who pays for advocacy — the identical provenance-hiding structure run at comparable scale by the left (Arabella), the right (DonorsTrust/Leo/Marble), and foreign state channels (USAID/NED), and how the same money wires into the designation machine.

The government cannot warrantlessly collect the data, so it buys the same data from private brokers instead — a purchase that launders the prohibition, sourced end to end.

Overwhelming tactical force pointed at people who did not warrant it — pre-dawn arrest theater on the non-violent, raids on journalists over newsgathering, and the innocent dead of mistaken-identity entries. Disproportion and wrong target, not ordinary drug or violent-crime raids.

How automation exposure concentrated in the lower-skill tier and the developing world produces a population to manage — and why the control is the programmable rail the transfer runs on, not the transfer itself.

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.

AI safety teams dissolved when they conflicted with product velocity; surveillance vendors selling to authoritarian regimes; security vendors captured or owned by state actors. Routed from the security-charlatans corpus.

The dominant form of great-power conflict is now non-kinetic: currency chokepoints, sanctions, tariffs, export controls, industrial policy, and — at the edge — designation-authorized force. The instruments, the competing state ratchets (US / China / BRICS), and why exercising a capability builds the counter-capability.

The control grid gets built everywhere, but it meets different resistance — and the variable is constitutional tradition, not culture. American entrenched rights block the speech ratchet and externalize control; British parliamentary sovereignty enables it; French general-will republicanism justifies it. Three shapes of control, not a freedom leaderboard.

A public-health problem was reclassified as an armed conflict, and the reclassification unlocked lethal force without trial. FTO designations, an 'unlawful combatant' finding, a withheld legal memo, 210+ dead at sea, the invasion of Venezuela and capture of its president, and an expanded domestic removal power — the narcoterror pawl, documented.

The AI-governance debate is a fight over who operates the control grid, not whether one gets built. Hard law, voluntary standards, and compute chokepoints — and what held versus what reversed in 2025–26.

A magazine Peter Thiel funded, except he didn't, because it never existed, because a language model invented it and handed it to us with a straight face. We used the machine to fact-check the machine, counted forty fabrications, learned they cluster exactly where nobody can check — and then caught our own cleanup committing the identical error: deleting a real company as a fake because we couldn't find it. Absence of evidence, laundered both directions.

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.

How the world's default encyclopedia became a contestable governance surface — read off its own record, with the self-correction that complicates the story left in.

The documented Chinese control-infrastructure record — social credit, the CCTV state, the surveillance majors, Xinjiang, e-CNY, and the export footprint — sourced to primary and named-investigative work, as the parallel to the US/UK body-control stack.

The receipts for the body layer of the control grid — biometric ID, identity-tied CBDC, brain-computer interfaces, genetic databases, programmable molecular platforms, and continuous monitoring — each advancing program paired with its own documented limit.

How OpenAI's pre-2024 exit paperwork tied a lifetime non-disparagement clause to vested equity — the mechanism, the reform arc, the refusers, and how the practice compares across labs and industries.

How a reproductive trajectory once private to adults became a routed transaction across a vendor stack — algorithmic child-welfare scoring, polygenic embryo screening, genetic-database property fights, biometric identity gating, clinical AI selection, and the 'good genes' rhetoric around it — each layer sourced, with contested efficacy claims kept attributed to the company that makes them.

The empirical record that AI does not merely offload cognition but actively steers it — persuading, completing, flattering, ranking, and bonding — with the ceiling on the claim documented alongside it.

AI-mediated employment decisions where a documented human harm sits in a court record, regulator action, or named principal's own words — the hiring screeners, video-interview scorers, firing algorithms, and the case law now holding the vendors liable.

The people who left, warned, refused, or were destroyed by the AI apparatus — each adverse claim attributed to a court, a filing, a sworn hearing, or the person's own on-record words, and no cause adjudicated that a tribunal has not.

How the Chapter 11 bankruptcy of 23andMe transferred roughly 15 million customer genetic profiles to a founder-controlled nonprofit through a court-supervised sale rather than individual re-consent.

The Behavioural Insights Team as a ratchet method — behavioral science moving from a Cabinet Office desk to a global consultancy, and 'nudge' sliding from opt-in choice architecture toward fear-based compliance engineering.

The tradecraft of subversion as a spectrum — from kinetic intelligence work (assassination, sabotage, honeypots) through active measures, ideological fronts, and Stasi Zersetzung to behavioral compliance and soft censorship. The ratchet prefers the soft tools: cheap, deniable, scalable, bloodless.

Separating documented fact from conspiracy — Charles III's real ties to Islam, Islamic finance, Perennialism, and Gulf-state donations, and why the 'Muslim Brotherhood' framing is an unsupported myth traceable to LaRouche-style conspiracy literature.

A symmetric research position for the Evil Robots series and evilrobots.lol

Left versus right is a rigged map: a narrow statist band with the exit painted over, and the one dimension that predicts the cage kept off the chart. Here are the three axes that actually sort who becomes a cage — grounding, method, reversibility — read off documented action, not labels.

A research position. Sourced, present-all-sides, and careful to separate the law that exists from the law that is merely wished for — the electrode is regulated; the feed is not.

A research position. Measured, sourced, present-all-sides. Peer-reviewed work is

A research position on the AI-worldview camps — doomers, accelerationists, present-harms, realists, refusers, dignity. Measured, present-all-sides, every claim attributed to the camp's own words.

US aid to Israel as the official, appropriated, statutory record — the 2016 MOU, FMF, missile-defense co-funding, the QME statute, loan guarantees, and the 2024 supplemental — sourced to CRS, public laws, State, and GAO.

The pro-Israel lobby in US politics as a coalition of named organizations, donors, and FEC-documented political spending — same documentary standard as the Cuban, Armenian, Indian, and Chinese influence apparatuses

Source dossier for The Ratchet (Book 2) -- Ch07 'The Priest' and Ch20 'The

Documented Israeli state-linked information operations — STOIC/Zero Zeno, institutional hasbara, the Oracle/TikTok question — held to the same standard as Russia's IRA and China's UFWD, and bounded the same way.

A documented legal mechanism — the Law of Return plus the 1978 non-extradition rule — producing named cases, surfaced and quantified almost entirely by Jewish outlets and Jewish child-protection advocates.

Living document. Index of named DOJ/OLC/judiciary and Trust-and-Safety personnel, their institutional rotations, and the primary receipts behind each — the legal-establishment layer of the control grid.

The political-economy of citizen rage on US immigration/labor-arbitrage discourse. Who collects the rent — media outlets, advocacy 501(c)s, politicians, platforms — and at what scale, per primary-source financial filings. Companion to the six-dossier diaspora/labor-arbitrage workstream in The Ratchet.

The discourse layer wrapping the labor-arbitrage / diaspora-influence / immigration mechanism — recurring frames, talking points, and the orgs / spokespersons / media outlets that deploy them, in the US and peer liberal democracies. Companion to six in-flight mechanism dossiers (US PACs, H-1B body shops, EU/UK, CA+AU, postwar intel, historical precedents). The Ratchet needs this to show how the mythology obscures the institutional capture documented in the other six.

Why Mexico runs no AIPAC/USINPAC-style diaspora PAC — not absence from the influence game but the opposite: the Mexican state is the actor itself, backed by a remittance stream that exceeds its oil revenue and a cartel export economy Washington treats as a national-security threat.

Pre-1900 historical cases of foreign-commercial-dominance displacing or threatening native economic and political position. Deeper-dive dossier expanding Section 7 of [Historical Precedents — When Polities Confronted Foreign-State or Foreign-Commercial Displacement](/research/research-historical-precedents-foreign-influence/), supplying the Embassy / Quarter chapter argument with case-by-case historical grounding. Chronologically organized, ancient-through-19th-century, with primary source where available, named historians per case, and an adversarial-review appendix that pre-empts both libertarian and restrictionist cherry-picks.

Snapshot date: 2026-05-29. Dataset state at completion: 373 named persons / 297 institutions / 789 edges.

Intelligence services as institutional actors instrumentalizing diaspora communities — postwar Mossad/CIA/KGB template and its worldwide adoption by MSS, RAW, MOIS, MIT, FSB/GRU, GIP, and others

Historical precedents and responses across the catastrophic-to-tolerable spectrum when polities faced foreign-state-influence or foreign-commercial-dominance displacement of native economic/political position. Sixth companion dossier to the diaspora-influence research workstream.

The Health-Governance Ratchet is structurally distinct from the US/UK revolving-door Ratchets. It operates via philanthropic funding determinism + institutional embedding, not personnel circulation.

H-1B/L-1 labor-arbitrage and discrimination enforcement record — DOJ settlements, congressional findings, top-employer concentration, wage-suppression studies. Companion to Embassy/diaspora-PAC dossier in The Ratchet ch 8.

European analog to the US diaspora-influence + labor-arbitrage Ratchet pattern — per-country diaspora-influence operations (UK, DE, FR, NL, SE, BE, CH, Eastern Europe) and EU/national labor-arbitrage mechanisms (Posted Workers Directive, UK Skilled Worker / Health & Care Worker visa, NHS recruitment, Blue Card / Chancenkarte, Schengen freedom of movement, informal southern-EU sectors). Companion to two in-flight US dossiers (diaspora PACs + H-1B body shops). Structural argument, sourced to parliamentary records, court rulings, regulator findings, and major-outlet investigations. No motive imputation. UK-libel-aware.

Diaspora political-action committees and advocacy orgs as remittance-economy influence channels in US policy

Tier 1: Finance-led coordination

Canada and Australia Ratchet pattern — Five Eyes companions to the US diaspora-PAC + labor-arbitrage dossiers, anchored on Hogue Commission (Canada) and FITS register (Australia) public records

Compiled May 2026. Primary sources cited throughout. Secondary sources flagged with outlet and date.

Compiled May 2026. Primary sources cited throughout. Secondary sources flagged.

The Twitter Files, all 19 installments: the FBI relationship, Hamilton 68, the Virality Project, and the government-to-platform flagging channel — with primary receipts and the downstream litigation.

January 6: Capitol riot. January 8-9: Google removes from Play Store, Apple from App Store. January 10 (midnight): AWS terminates hosting — Parler fully offline. January 11: Parler sues Amazon. January 21: the court denies the injunction.

Chinese state-sponsored hackers compromised at least nine US telecom companies: AT&T, Verizon, T-Mobile, Lumen, Spectrum, Consolidated Communications, Windstream. Specifically targeted the CALEA-mandated lawful-intercept systems — the wiretapping backdoors the US government required telecoms to build.

Reference document for institutional control infrastructure research. Covers workplace algorithmic management, employee monitoring software, and the consumer IoT surveillance ecosystem.

The foundation layer. Everything in the modern AI control grid is built on top of this.

The nation-state breach layer: Anthem, Change Healthcare, Equifax, Marriott, OPM — bulk health and identity data harvested at scale, attributed to named state actors (MSS, PLA, APT41), and why the genome became strategic intelligence.

The thesis: every surveillance capability governments build eventually gets turned against them.

Flo Health FTC settlement: Shared data from millions of users with Facebook, Google despite privacy promises. Didn't stop until 2019 WSJ exposé. FTC finalized order June 2021. In 2025, Google paid $48M, Flo $8M, and Flurry $3.5M (total $59.5M).

What they monitor: Webcam, audio, screen, keystroke patterns, eye tracking, room scans. ExamSoft collects 'retina or iris scan, fingerprint, voiceprint, or scan of hand or face geometry.'

1.4 billion enrollments. World's largest biometric identity database — fingerprints, iris scans, facial photographs linked to 12-digit ID.

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.

Ivanski, Lo & Mar (2021): conservatism predicts disliking cats and preferring dogs; RWA (not SDO) mediates the cat aversion. The psychology base for the Cat-or-Dog quiz.

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 girls' mental health, and the ranking system amplifies outrage because engagement rewards it.

Automated detection accounts for over 90% of content actions. In 2021, Meta claimed AI proactively removed 97% of hate speech before anyone reported it.