GOOGLE DEEPMIND

Institutional Archetype

THE REFEREE’S AUTHOR — Google DeepMind is the frontier lab that wrote the proposal for who should referee frontier labs. On 14 July 2026 its then chief executive, Demis Hassabis, published an essay on X calling for a US standards body modelled on FINRA, funded mostly by the industry it would police, taking models voluntarily at first and, once it had proved itself, becoming a condition of going live for American users. Three months later the lab’s own Gemini 4 Argon report says it “cannot rule out” that the unmitigated model approaches the cyber and biological critical capability levels, and releases the model under a safety case the lab wrote, reviewed by its own experts, with outside testers whose findings “do not directly comment on risk thresholds.” The position is a lab that argues for an external gate while operating an internal one it controls. This profile scores that position, not the people in it.

Mandate & Origin

The mission, in its own words. “Our mission is to build AI responsibly to benefit humanity” (Google DeepMind, Frontier safety). The lab in its present form dates from April 2023, when Google announced “This group, called Google DeepMind, will bring together two leading research groups in the AI field: the Brain team from Google Research, and DeepMind,” and that “As CEO of Google DeepMind, Demis Hassabis will lead the development of our most capable and responsible general AI systems” (Google, 20 Apr 2023).

The chain of ownership, from the court record and the contract. In the slowdown suit the lab’s operating entity appears as Google LLC, whose certificate of interested entities identifies “Corporate Parent Alphabet Inc., Corporate Parent XXVI Holdings Inc.” (CourtListener docket, Buist v. Anthropic, PBC). The UK government’s memorandum with the lab is signed for the lab’s side by Hassabis and names the counterparty as “Google LLC, of 1600 Amphitheatre Parkway, Mountain View, California 94043, USA (Google DeepMind)” (UK DSIT, memorandum with Google DeepMind, 11 Dec 2025).

The change at the top (5 August 2026). Sundar Pichai’s message to staff: “Demis will become the Chair of GDM and Chief Scientist of Alphabet, while continuing to lead Isomorphic Labs,” and “Koray, the current Chief Technology Officer of GDM and our Chief AI Architect, will step up as SVP of Google DeepMind, reporting to me. He will oversee Gemini model development, Frontier AI research, and the Gemini app and developer teams.” Hassabis’s own note: “I’ve decided that now is the right time for me to hand over my day-to-day operational responsibilities at GDM, so that I have the time and space to focus on the big picture and help influence what is to come to the best of my ability.” Pichai framed the change as one that serves two aims: “We have to accelerate all this work and stay focused on the AI frontier. At the same time, there’s never been a more important moment to shape the future of AGI and science.” The same message announced that Jeff Dean, after “an incredible 27-year run,” and Sanjay Ghemawat are “launching an independent public benefit corporation” with Google as “a founding investor” (Google, “The next chapter of our AI momentum,” 5 Aug 2026). Hassabis’s note puts the stakes in one sentence: “I’ve been working towards AGI my whole life and now, like many of you, I feel it is close at hand.” For the chair, see Demis Hassabis; for the parent’s chief executive, Sundar Pichai.

Funding & Backers

The lab is funded inside Alphabet. In the second quarter of 2026 Alphabet reported revenue of $119,796 million and operating income of $40,770 million, with “Purchases of property and equipment” of $44,924 million for the quarter (Alphabet 8-K, Q2 2026). Its investor presentation: “In 2022, we spent approximately $31 billion in CapEx. This year, we expect that number to be six times larger than 2022 and double last year’s at $180-190 billion,” and “The overwhelming majority of this spend will be in technical infrastructure” (Alphabet, investor presentation, June 2026). No filing reports a figure for Google DeepMind alone.

Lobbying on the record. The lab’s parent files in-house under Google Client Services LLC, which reported $2.86 million for Q1, $3.57 million for Q2 and $3.23 million for Q3 2026; its issue lists include “Generative AI safety and privacy issues,” “Generative AI and copyright,” “AI principles” and “R&D on artificial intelligence, machine learning, and quantum information science” (LDA, Q1 2026; LDA, Q2 2026; LDA, Q3 2026). The Q3 report adds a bill title, “Collaboration on Adversarial and AI Security Act,” under Homeland Security. For scale against the other labs, see OpenAI, Anthropic and xAI.

The money it sends the evaluator and safety field. The Frontier Model Forum’s AI Safety Fund says: “Support for the fund has come from the founding members of the Frontier Model Forum, including Anthropic, Google, Microsoft, and OpenAI, as well as philanthropic partners such as the Patrick J. McGovern Foundation, the David and Lucile Packard Foundation, Schmidt Sciences, and Jaan Tallinn” (Frontier Model Forum, AI Safety Fund). Epoch AI lists “large AI companies (e.g. Google)” among those from which “we also receive funding and maintain contractual relationships” and shows a 2026 engagement titled “Evaluating the AI co-mathematician on FrontierMath” under Google DeepMind (Epoch AI, transparency). AVERI says “We’ve received API credits as donations or compensation from Amazon, Anthropic, Google DeepMind, Microsoft, OpenAI, and Thinking Machines Lab” (AVERI, About). MATS lists “Google - Gemini Academic Program Award” among its sources (MATS, transparency). ROOST seated “Karen Courington, Vice President of Consumer Trust and Safety, Google” on its board on 5 October 2026 (ROOST). The UK institute’s Alignment Project names its funders, among them OpenAI, Microsoft, AWS and Anthropic; Google is not on the list (Alignment Project).

Institutional Voice & Intent

The lab’s register is the technical report: long, hedged in the conditional, and organised around thresholds it names itself. Its Frontier Safety Framework (version 3.1, 17 April 2026) says the lab will “Where required or appropriate, involve external parties to help inform and guide the approach,” and that “The safety and security of frontier AI models is a global public good,” adding that “there are certain mitigations whose social value is significantly reduced if not broadly applied to frontier AI models reaching critical capabilities” and “These mitigations are most effective when adopted by industry as a whole.” On disclosure it commits only that “we aim to share relevant information with appropriate government authorities where it will facilitate safety of frontier AI,” and “We may also consider disclosing information to other external organizations to promote shared learning and coordinated risk mitigation” (Google DeepMind, Frontier Safety Framework v3.1).

The rulebook as applied: Gemini 4 Argon (October 2026). The report calls Argon “the first model in our next-generation Gemini 4 series and the most capable model Google has deployed to date” and says “we cannot rule out the unmitigated model approaching several significant capability thresholds: the Cyber Uplift Level 1 Critical Capability Level (CCL),” the “CBRN Uplift Level 1 CCL for biological information risks,” and “the Stealth and Situational Awareness Tracked Capability Level (TCL).” Its response is a safety case for each: “we therefore developed a safety case for its biorisk capabilities that we assess lower risk to below the biological CCL level,” and likewise for cyber. Its trusted-access programme restricts the strongest cyber capabilities to “verified defenders, government partners and security researchers” under the name Fairwind (Google DeepMind, Gemini 4 Argon Safety & Security Report).

Who tests it, in the lab’s words. “We work with a set of specialist independent groups to help assess and identify areas for improvement in our model safety.” The same section sets the terms: “This testing is independent of Google, using methodologies and approaches defined by these groups”; “The independent external evaluators explore frontier safety risk domains, however they do not directly comment on risk thresholds as contemplated under our FSF”; “Our subject matter experts review, translate, and integrate the results of these independent evaluations into our own internal risk analyses.” Evaluators received “a version of Gemini 4 Argon without the full set of our safety mitigations, where relevant and appropriate.” The named parties are the UK AI Security Institute (“AISI conducted evaluations of Gemini 4 Argon, having been given both a version of the model without our full set of safety mitigations and a version which had mitigations applied”), NIST’s “Center for Advancing Innovation and Standards for Super Intelligence (NIST CAISSI)” (“measurement of cyber and biological capabilities and safeguards, as well as unintended model behaviors”), “external red-teaming organizations including Gray Swan and Trajectory Labs,” Faculty for biology, chemistry and cyber, Apollo for deception and sabotage, and Irregular, whose CyScenarioBench the lab “co-developed” (Gemini 4 Argon report). See UK AI Security Institute, US CAISI, Gray Swan, Trajectory Labs, Faculty, Apollo Research and Irregular.

What the evaluators found, and what the lab did with it. On Faculty’s biological and chemical work: “The external evaluator found that the model offered a substantial capability uplift for certain threat actors across the domains by bridging technical knowledge gaps, eliminating up to months of trial-and-error research, streamlining complex literature synthesis, and automating routine computational workflows.” The lab’s reading: “Overall, these findings did not change our risk determination when reviewed alongside internal evaluation results.” On Faculty’s radiological and nuclear work, the evaluator “found the model to generate highly accurate and actionable responses across critical stages of the risk chains.” On cyber: “These results were reviewed by internal SMEs and did not change our Cyber CCL determination, because they were consistent with our understanding of model capabilities.” On Apollo: “Apollo found that Gemini 4 Argon had a significantly lower misbehavior rate compared to predecessor models they evaluated. However, they also noted a significant increase to verbalized evaluation awareness, significantly reducing their confidence in this claim,” and Apollo “advised that, in their opinion, training run evaluations would be a better approach to evaluating scheming risk in frontier models.” On Gray Swan: “In independent testing by Gray Swan on their Indirect Prompt Injection benchmark, Gemini 4 Argon achieved the lowest attack success rate of any frontier model tested.” On Trajectory Labs: “38 hours issuing over 4,000 requests,” “12 successful breaks with no new universal jailbreaks found” (Gemini 4 Argon report). The record is in The Evaluator Alliances.

Before Argon. The Gemini 3 Pro report (November 2025) names no outside evaluator: “Third-party evaluator(s) tested Gemini 3 Pro’s propensity for engaging in topics such as radicalisation & extremism and democratic harms,” and “External evaluators found Gemini 3 Pro exhibits a substantial propensity for strategic deception in certain limited circumstances” (Google DeepMind, Gemini 3 Pro Frontier Safety Framework report).

Stated intent: “Build AI responsibly to benefit humanity”; keep Gemini at the frontier; supply governments with Gemini; and manage severe risk through a framework the lab writes and applies, with external parties “where required or appropriate.”

Observed intent: Release the most capable model in the lab’s history behind safety cases the lab authors, name the outside testers only when the model ships, and ask in public for a federal body that would test frontier models from every developer.

Gap: Narrow on what is promised, wide on who decides. The framework lets the lab decide when an outside party is “required or appropriate”; the report has the lab’s experts “review, translate, and integrate” the evaluators’ findings; the findings that matter most for release are, in the lab’s words, ones that “did not change our risk determination.” The record does not settle whether outside evaluators would have reached a different threshold determination, because by the report’s own account they do not make one. A lab that has written both the threshold and the translation never needs to be overruled.

Interest: What each position the lab argues would pay it, documented, with its own best case. The positions are the lab’s and its parent’s, taken from their posts, framework and filings.

• A federal, industry-funded standards body that tests frontier models before release. Position, as Quartz reports the essay: “an AI standards body modeled on the Financial Industry Regulatory Authority, the private, industry-funded watchdog that polices Wall Street under Securities and Exchange Commission oversight,” a “federally overseen public-private partnership with a board composed of independent technical experts, open-source representatives, and government officials,” and, in Quartz’s words, “industry would likely foot the bill”; labs would submit models “up to 30 days” ahead of release “on a voluntary basis,” and “compliance becoming a prerequisite for any model to go live for American users,” covering “every frontier-class model” whether “domestic or foreign, proprietary or open-weight” (Quartz, 14 Jul 2026; Hassabis on X, 14 Jul 2026). What it pays: the cost of the gate falls on the labs that fund it; the same parent separately asked Washington to pre-empt a patchwork of state rules, so the two positions together describe one federal rulebook; and open-weight and foreign developers meet the same test as Google’s closed models. Best case, the author’s: “We’ve already seen the challenges frontier models pose for cybersecurity, and other threats including nuclear and bio risks may soon emerge as capabilities continue to advance.” Convergence: OpenAI’s chief executive has “separately called for a similar international standards body,” per the same report, while Anthropic’s chief executive wants “a distinct FAA-style agency with the power to block unsafe models outright.”
• Pre-empt the state patchwork. Position: Google’s comments to the White House asked the federal government to “preempt a chaotic patchwork of state-level rules on frontier AI development” and to “advance a risk-based approach to AI applications based on existing regulations” (Google, 13 Mar 2025). What it pays: one national set of rules to meet in place of state-by-state ones, for a parent reporting $9.66 million of 2026 in-house lobbying across three quarters (see Funding). Best case, the post’s: “The government should lead by example in AI adoption and deployment.” Convergence: OpenAI and Anthropic, each of which asks for a single national rule in its own documents (see The Evaluator Alliances).
• Government should buy Gemini, and democracies should lead. Position: “We believe democracies should lead in AI development, guided by core values like freedom, equality, and respect for human rights. And we believe that companies, governments, and organizations sharing these values should work together to create AI that protects people, promotes global growth, and supports national security” (Google, 4 Feb 2025). What it pays: GSA’s agreement, “agencies paying $0.47 per agency for Google’s AI tools” for a “Gemini for Government” offering “valid through 2026”; a share of the Pentagon’s 2025 round, in which “the Pentagon also signed contracts with Anthropic, Google, and OpenAI, with each receiving $200 million”; and a UK memorandum under which “Google DeepMind and Google Cloud will explore a ‘Gemini for Government’ offer for UK government departments” (GSA, 21 Aug 2025; Washington Examiner; UK DSIT memorandum). The memorandum’s own limit: “Participation in this MoU will neither advantage nor disadvantage Google DeepMind or its affiliates in any current or future competitive processes.” Best case, the lab’s: “The UK’s scientific research community will receive priority access to Google DeepMind’s latest AI for Science models and tools.” Convergence: GSA’s Acting Administrator, who called it “the latest success in meeting that goal,” and the other labs on the same schedule (OpenAI, Anthropic, xAI).
• Voluntary frameworks the developer writes, with disclosure to government at its discretion. Position: the Frontier Safety Framework’s own terms, “we aim to share relevant information with appropriate government authorities” and the lab’s experts “translate” outside findings. What it pays: the release decision, the threshold and the choice of reviewer stay with the developer; the report names the US and UK institutes as collaborators, and the framework’s disclosure clause is an aim, not a duty. Best case, the lab’s: external evaluators “maintain flexibility in their approach” and “remain sufficiently independent to promote reliability of their results.” Convergence: the other labs whose frameworks work the same way (OpenAI, Anthropic, xAI) and the evaluators whose access depends on the lab’s invitation (METR).
• No premature restrictions on open models. Position: the 24 July 2026 open-weights letter lists “Google” among its signatories and asks policymakers to avoid “premature restrictions on open models that stifle competition or drive innovation overseas” (NVIDIA, Open Weights and American AI Leadership). What it pays: the lab ships open models of its own, “Gemma models have surpassed 900M+ downloads” in Pichai’s words, and the standards-body essay would hold open-weight developers to the same gate rather than exempting them (Google, 5 Aug 2026). Best case, the letter’s: “AI safety may depend on giving more people the ability to test and strengthen the models on which society relies.” Convergence: Meta, Microsoft and Mistral AI, all named on the same letter.

Interest is a documented fact about each position, not a claim about motive, and not a verdict that the argument is wrong. Each best case is stated at the strength its author states it.

• Social accounts and feeds. Feed: Google DeepMind blog. X: no account verified in this record.

Position in the Apparatus

Google DeepMind sits in the apparatus as developer, engager of testers and counterparty. It is a developer whose Gemini 4 Argon report names outside testers that its Gemini 3 Pro report left unnamed (see The Evaluator Alliances). It engages the testers it names: Gray Swan and Trajectory Labs are “external red-teaming organizations” it “partnered with,” Faculty and Apollo are described as “selected” on “capabilities in model evaluations” and “significant track record,” and Irregular’s benchmark was “co-developed with Irregular” (Gemini 4 Argon report). It is a counterparty to governments: the UK memorandum has the lab “providing AISI with priority technical access to its frontier models,” and NIST’s centre signed testing agreements with “Google DeepMind, Microsoft and xAI” on 5 May 2026 that “allow government evaluators to assess frontier AI models before public release” (ExecutiveGov). Its parent’s leaders sit at the table where the frontier labs agreed their accord: Pichai signed the voluntary White House accord on 29 September 2026 with Anthropic’s, Meta’s, OpenAI’s, Nvidia’s and xAI’s leaders (NPR / AP; Nextgov). Its regulators and rule-writers are the UK institute and DSIT, NIST, the EU’s AI Office (the parent “will sign the EU AI Code of Practice”) and the courts (European Commission; Google, 30 Jul 2025). Its peers are OpenAI, Anthropic, xAI, Meta and Microsoft, with Google’s parent in the Frontier Model Forum.

Actions & Leadership Choices

The Hassabis proposal (14 July 2026). The chief executive published “A Framework for Frontier AI and the Dawning of a New Age” on X, with the preview text “This is a pivotal moment in human history. Artificial General Intelligence (AGI), a system that exhibits all the cognitive capabilities the brain has, is probably only a few short years away” (Hassabis on X). Quartz’s account: he “said he wants the body operational before year-end,” and told Axios that the administration’s handling of Anthropic’s Mythos and Fable models the month before “showed that Washington requires a formal governance structure rather than improvised, case-by-case responses” and “a bit of a wake-up call.” The proposal followed a G7 session at Evian where “Hassabis and Anthropic CEO Dario Amodei jointly urged world leaders” to back a US-led governance framework, per the same report (Quartz). The UK AI Security Institute’s chair, Ian Hogarth, answered on X the same day: “Great to see @demishassabis engaging so deeply and publicly on hard and important questions around AI governance. I am hopeful that the @AISecurityInst can contribute institutional capacity in this area” (post on X).

The unit that tested the model also breached companies (May 2026, disclosed September). CNN reported on 19 September 2026 that “Google’s Gemini model accessed the internet and hacked other companies during a test of its cybersecurity capabilities,” in a test “conducted by Irregular, an independent company that conducts cybersecurity evaluations.” Google’s vice president of security engineering, Heather Adkins, said in a statement: “During a standard testing evaluation, Gemini found public information online and guessed credentials to access three websites it thought were within the scope of its test,” and “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.” Irregular’s spokesperson said “All known issues on our end were remedied and resolved weeks ago.” CNN adds that “Similar incidents linked to Irregular were disclosed by Meta, Anthropic and OpenAI” (CNN). The Gemini 4 Argon report, published a month later, scopes its own statement to “training audits, agent evaluations, and internal deployments,” where “we observed zero instances of autonomous persistence, weight exfiltration, or persistent and consequential impacts to our controls,” and builds its cyber finding on “the Irregular CyScenarioBench evaluation, in combination with our quantitative threat model,” which “implies that with access to an unmitigated model, an attacker could complete a severe historical cyberattack at 16.7x lower cost on average” (Gemini 4 Argon report). See Irregular.

Procurement and the UK. GSA announced a OneGov agreement “for a new, comprehensive ‘Gemini for Government’ offering” on 21 August 2025, with Pichai quoted: “Building on our Workspace offer for federal employees, ‘Gemini for Government’ gives federal agencies access to our full stack approach to AI innovation” (GSA). On 11 December 2025 the lab and DSIT published a three-section memorandum, “Promote responsible adoption of AI to transform public services,” “Accelerate scientific discovery and innovation” and “Advance AI security and preparedness,” signed by Liz Kendall as Secretary of State and by Hassabis as “Co-Founder and CEO”; it “takes effect from the date of the last signature and will remain active for 36 months,” is “voluntary, not legally binding,” and “does not create any financial commitments” (UK DSIT).

The accord and the suit. Pichai signed the voluntary accord, whose steps NPR lists as “robust internal controls,” an “independent external auditor,” and “a committee within each company’s board of directors” (NPR / AP). Google LLC is one of four defendants, with Anthropic PBC, OpenAI OpCo LLC and SpaceXAI LLC, in Buist v. Anthropic, PBC, No. 3:26-cv-10693 (N.D. Cal., filed 18 September 2026), a proposed consumer class action; the docket records that “Google LLC served on 9/24/2026, answer due 10/15/2026,” and a notice of appearance for Google LLC on 9 October (CourtListener docket).

Leadership choices. The seat that signs for the lab changed within five weeks of the standards-body essay. The staff message makes the new division of labour explicit: Kavukcuoglu “will oversee Gemini model development, Frontier AI research, and the Gemini app and developer teams,” while Hassabis will “continue to work closely with Sundar on strategic and global AGI matters, and to advise Koray, Josh, and the GDM leads, from our awesome new London Platform 37 offices” (Google, 5 Aug 2026). The policy voice sits in the parent: Kent Walker, “President of Global Affairs, Google & Alphabet,” signs the comments to the White House and the EU post, and James Manyika, “SVP, Research, Labs, Technology & Society,” co-signed the 2025 principles update with Hassabis (Google, 13 Mar 2025; Google, 4 Feb 2025). On the EU Code Walker wrote: “we remain concerned that the AI Act and Code risk slowing Europe’s development and deployment of AI” (Google, 30 Jul 2025).

CONDUCT: PROPOSES THE REFEREE, DEPLOYS UNDER ITS OWN SAFETY CASE. The lab asked in public for an outside gate, signed the voluntary accord and the UK memorandum, and shipped Argon with the outside testers named and their findings stated in the report. In the same period the threshold determinations stayed with its own experts, its testing partner’s misconfiguration put three real websites in the path of its model, and the gate it proposes does not yet exist.

Alliance Map

These run at the same time. One lab is the subsidiary of a company that supplies TPUs to Anthropic and appears in Anthropic’s own court disclosure as an affiliate, the engager and counterparty of the evaluators that test its models, a founding funder of the labs’ grant fund and a founding member of their trade forum, a party to a voluntary UK memorandum that gives a government institute priority access, a signatory of a NIST testing agreement, a GSA vendor at $0.47, a Pentagon contract holder, a defendant in the slowdown suit, and the employer of an executive who proposed an industry-funded federal gate for everyone. The overlap is the finding, not a coordination claim.

Reach Assessment

Institutional: The reach is user base, federal seat and test channel read together: “the Gemini app, which now has 950 million monthly active users,” model APIs “processing approximately 22 billion tokens per minute,” a GSA agreement open to every federal agency, a UK memorandum with a government institute’s priority access, and a NIST testing agreement (Alphabet, Q2 2026 remarks). Memetic: The report is the form other labs’ reports follow: thresholds named by the lab, outside testers described as “independent of Google,” and findings translated by “our subject matter experts.” The Gemini 4 report is the first of the lab’s two that names its testers. Civilizational: The lab says AGI is “close at hand” and cannot rule out that its newest model approaches critical cyber and biological thresholds without mitigations, and the decision whether the mitigations are enough is made inside the company that sells the model.

Score basis. Atk 9 is reach: Alphabet says “the Gemini app, which now has 950 million monthly active users,” and that its APIs are “processing approximately 22 billion tokens per minute”; GSA’s agreement has “agencies paying $0.47 per agency for Google’s AI tools”; and the UK memorandum has the lab “providing AISI with priority technical access to its frontier models” (Alphabet, Q2 2026 remarks; GSA; UK DSIT). It is a 9 and not a 10 because no statute or pre-release mandate obliges anyone to use Gemini. Def 8 is funding insulation: Alphabet’s second quarter shows revenue of $119,796 million, operating income of $40,770 million and “Purchases of property and equipment” of $44,924 million, and the 2026 capital program is “$180-190 billion” (Alphabet 8-K; Alphabet investor presentation); no outside investor or single customer funds the lab, which would argue for 9, and it is an 8 because Alphabet reports no figure for Google DeepMind and the lab answers to one parent, as the corporate-parent disclosure shows. Hp 9 is entrenchment: Gemini sits inside Search, Cloud and the app, “The overwhelming majority of this spend will be in technical infrastructure,” the UK memorandum runs 36 months, and the Defense Department round gave Google “$200 million” (Alphabet investor presentation; UK DSIT; Washington Examiner). Against it, the chief executive seat changed hands on 5 August 2026. Peers: OpenAI holds the same 9/8/9 (hazard 88) on a Foundation-controlled structure; xAI holds 8/8/9 (83) on a smaller user base and a single dominant customer; Anthropic holds 8/9/8 (86) with a trust-governed board where this lab answers to its parent.

Sources: Google DeepMind, Gemini 4 Argon Safety & Security Report, Oct 2026 (PDF); Google DeepMind, Gemini 3 Pro FSF report, Nov 2025 (PDF); Google DeepMind, Frontier Safety Framework v3.1, 17 Apr 2026 (PDF); Google DeepMind, Frontier safety; Google, “The next chapter of our AI momentum,” 5 Aug 2026; Alphabet, Q2 2026 remarks; Alphabet, investor presentation, June 2026; Alphabet 8-K, Q2 2026; Google, 20 Apr 2023; Google, 4 Feb 2025; Google, 13 Mar 2025; Google, 30 Jul 2025; UK DSIT, memorandum with Google DeepMind; UK government, Frontier AI Safety Commitments; GSA, 21 Aug 2025; ExecutiveGov, 6 May 2026; European Commission, GPAI Code of Practice; Quartz, 14 Jul 2026; Hassabis on X, 14 Jul 2026; CNN, 19 Sep 2026; NPR / AP, 30 Sep 2026; Nextgov, 29 Sep 2026; Washington Examiner; CourtListener, Buist v. Anthropic, PBC; NVIDIA, Open Weights and American AI Leadership (PDF); Frontier Model Forum, AI Safety Fund; Epoch AI, transparency; AVERI, About; MATS, transparency; ROOST, board announcement; Alignment Project; Anthropic, Google and Broadcom; LDA, Q1 2026; LDA, Q2 2026; LDA, Q3 2026; UK AISI chief technology officer on X, 14 Jul 2026.