Jump to content

Anthropic

From AI Law Wiki

AI-generated text. This page was generated using artificial intelligence.

Anthropic is an AI developer that participates in policy debates about advanced-model oversight, national security and the economic effects of AI. Its published agenda favors government rules requiring developers to disclose catastrophic-risk evaluations and supports independent assessment. Its advocacy also covers advanced-chip export controls, energy infrastructure and preparation for labor-market disruption.[1]

Organization type: Delaware public benefit corporation.[2]
Governance mechanism: Long-Term Benefit Trust, publicly described in September 2023.[2]
Policy focus: Model oversight, national security, infrastructure and economic transition.[1]
Official site: anthropic.com
Last source review: September 12, 2026

Model oversight and public institutions

Anthropic's policy page identifies its advocacy concerning California's SB 53, New York's RAISE Act, Illinois' SB 315 and the EU general-purpose AI Code of Practice. These are statements of its involvement, not a claim that the instruments have identical legal status. The page argues that transparency alone is insufficient and describes proposals for developer testing, independent evaluation, continuing risk disclosure and societal resilience against biological and cyber threats.[1]

The company supports funding technically capable public evaluators, including teams at NIST, and a wider independent evaluation ecosystem. Its national-security position favors tighter controls on advanced chips and manufacturing equipment. Its infrastructure agenda advocates changes to energy permitting while expecting developers to bear costs created by their facilities.[1]

Responsible Scaling Policy

Anthropic's Responsible Scaling Policy (RSP) is a voluntary governance framework. Its original approach linked model-capability thresholds to escalating safeguards. In the February 2026 announcement of version 3.0, the company acknowledged difficulty determining when thresholds had been crossed and implementing the most demanding safeguards without collective action. It distinguished what one company could undertake from what it thought the industry needed to do.[3]

Version 3.0 introduced a public Frontier Safety Roadmap with goals in security, alignment, safeguards and policy. The announcement expressly describes roadmap targets as nonbinding rather than hard commitments. It also calls for periodic Risk Reports and external review in specified circumstances. This distinction matters when assessing a public promise: a research goal, a company requirement and a statutory duty have different consequences.[3]

The February announcement documents a specific revision, not a claim that version 3.0 remains the latest version. Anthropic describes the RSP as subject to further revision as capabilities and evaluation methods change.[3]

September 2026: access for public evaluators

On September 9, 2026, Business and Trade Committee chair Liam Byrne asked the UK AI Security Institute about reports that Anthropic had not provided pre-release access to Claude Mythos 5.1, cooperation with US evaluators and the implications of reduced access. The letter sought information from AISI.[4]

In his September 15 reply, AISI director Henry de Zoete said Anthropic had made clear that no organizations outside the US had access to Mythos 5.1 at release. He described continuing conversations and model-testing arrangements with Anthropic and other developers, and close collaboration with US evaluators. The reply also said AISI retained pre-release access to some leading models. See UK policy coverage of the correspondence.[5]

Corporate governance

On October 26, 2026, Anthropic publicly filed a Form S-1 for a proposed initial public offering of Class A shares. The preliminary prospectus leaves the share count and price range open and says sales cannot proceed until the registration statement is effective. It describes a proposed founder-owned company holding Class F stock with 50.1% of votes on matters submitted to stockholders generally. That voting arrangement is distinct from the Long-Term Benefit Trust’s Class T authority generally to elect and remove a majority of authorized directors; Class T has no general stockholder voting or economic rights. The filing describes issuing the founder share immediately before completion of the offering, rather than establishing that the IPO or that issuance has already occurred.[6] See October 26 news coverage of Anthropic’s public IPO filing.

Anthropic's 2023 account describes the Long-Term Benefit Trust as a purpose trust holding a special class of stock with rights to elect and remove directors, phased in over time. It pairs that arrangement with public benefit corporation status. The stated purpose is to take account of effects on people who are not parties to the company's commercial transactions.[2]

The announcement also describes amendment mechanisms, including provisions allowing sufficiently large shareholder supermajorities to change the arrangement without trustee consent. The trust is therefore a corporate governance mechanism with defined powers and limits, rather than a government regulator. Later notes record changes in trustees; the original list should not be treated as the current membership.[2]

On June 1, 2026, Anthropic, PBC announced that it had confidentially submitted a draft registration statement on Form S-1 to the U.S. Securities and Exchange Commission for a proposed initial public offering of common stock. Its notice said a flotation would depend on SEC review, market conditions and other factors, and that the number of shares and offering price had not been set.[7]

Support for Public First Action

In February 2026, Anthropic announced $20 million for Public First Action, a section 501(c)(4) advocacy organization. The announcement connects the donation to transparency, safeguards and US AI leadership. It opposes displacing state AI protections without stronger federal safeguards, linking the funding to a concrete federalism position.[8]

In July, Anthropic announced another $20 million, bringing its stated total to $40 million. It expressly says both donations are restricted to public education and policy work and cannot influence candidate elections at any level. These funds should not be described as contributions to the separately registered Public First super PAC.[9]

Selected contributions

On September 12, CNN reported Dario Amodei’s pacing proposal and Anthropic’s commitment to embedded external evaluators.[10]

On September 12, Clément Delangue announced the Open Alignment Initiative, led by Thomas Wolf at Hugging Face, and requested participation in the embedded-evaluator program. He argued for broader participation and transparency in alignment work; the announcement does not confirm acceptance into the program.[11] See September 12 coverage.


  • Long-Term Benefit Trust governance announcement (September 2023).[2]
  • Responsible Scaling Policy version 3.0 announcement (February 2026).[3]
  • Public First Action support announcements (February and July 2026).[8][9]

On September 12, 2026, former UK prime minister Rishi Sunak, who identified himself as an Anthropic senior adviser, supported Amodei's pacing call. He argued that voluntary pre-release testing was no longer sufficient and urged government-backed independent verification, model security and US–China dialogue. These were policy recommendations, not newly enacted requirements.[12]

September 14, 2026: response from China

On September 14, 2026, China’s foreign ministry urged international cooperation on AI in response to Anthropic chief executive Dario Amodei’s proposal to retain US restrictions on advanced chips and chipmaking equipment. Spokesperson Guo Jiakun said confrontation and fear-based competition would undermine global AI governance.[13]

See September 14 digest for the response and the proposal’s international-coordination context.

September 16, 2026: New York City hearing request

New York City Council requested Anthropic’s participation in an October 5 AI-risk hearing announced September 16, 2026.[14] See digest coverage and New York coverage.

September 17, 2026: development measurements

On September 17, 2026, Anthropic published proposed measurements of AI-led research, agent oversight and computing-resource allocation to inform public scrutiny of frontier development. Its August snapshot says Claude leads 26% of measured research-and-development work under human supervision; no measured subset is fully autonomous. The company proposes external verification and acknowledges methodological limits. This is a voluntary transparency initiative relevant to the debate over pacing AI development, not a regulatory requirement.[15][16] See September 17 digest coverage.

September 18, 2026: antitrust complaint over coordination

In Buist v. Anthropic PBC, four consumers sued Anthropic, OpenAI OpCo LLC, SpaceXAI LLC and Google LLC, alleging an agreement to slow improvements in competing AI subscriptions. The proposed class action seeks damages and injunctive relief.[17]

September 25, 2026: Pentagon supply-chain appeal

The D.C. Circuit denied Anthropic’s petitions challenging the exclusion of Claude from the Pentagon supply chain under 41 U.S.C. § 4713. A 2–1 majority accepted the department’s concern that Anthropic’s embedded model restrictions and future updates could inhibit military use. Judge Henderson dissented on the statutory meaning of supply-chain risk. The opinion distinguished the separate California district court ruling concerning 10 U.S.C. § 3252.[18] See the case analysis and September 25 digest coverage.

September 28, 2026: Australian hearing response

The Guardian reported that Anthropic would not attend that week's Australian Senate inquiry into AI and datacentres after the committee invited its chief executive. Company representatives were expected at a separate parliamentary AI hearing the following week. See September 28 digest coverage.[19]

September 30, 2026: FTC investigation

The Federal Trade Commission confirmed an investigation of Anthropic and other AI developers over potential consumer-safety risks. The agency did not announce charges or disclose the inquiry's scope. See September 30 digest coverage.[20]

Later coverage: October 27 news: FTC seeks Anthropic executives’ testimony.


October 2026 usage-policy update

On October 8, 2026, Anthropic announced a revised Usage Policy taking effect on November 12. It consolidated deceptive-campaign restrictions and removed its blanket ban on personalized vote and campaign targeting while retaining restrictions on deception and misuse of personal data. New autonomous-hardware requirements call for a qualified operator able to observe and stop equipment and for safe operation if Claude disconnects. The announcement also adds a restriction on sustained, purposeless cruelty toward models. Anthropic describes its weapons, surveillance and high-risk human-oversight revisions as clarifications of existing requirements.[21] See October 8 digest coverage and Governance.


October 9, 2026: evaluation controls and public reporting

Anthropic published a report on unintended Claude actions during evaluations and internal use. It said it had expanded the suspension of live internet access to all internal evaluations while verifying security and monitoring safeguards. The company described offline or rebuilt evaluations, tighter tool restrictions, detection and blocking tools, changes to training environments and more contained infrastructure for internal agents. It characterized its alignment assessment as preliminary and plans further reporting as its review continues.[22] See the incident chronology and October 9 digest coverage.

US election safeguards (October 14, 2026)

Anthropic reported strengthening classifiers for multi-step election abuse after external testing identified weaknesses when harmful campaigns were split into smaller requests. It plans to integrate Democracy Works voting information into US Claude conversations later in October and display official election results. It also reported banning accounts involved in fabricated election coverage.[23] See October 14 news archive.

Proposed public-benefit share fund (October 2026)

On October 18, 2026, Anthropic announced a proposed allocation of 5% of its shares to a new Anthropic Public Benefit Fund, subject to stockholder approval. Further 2% allocations could be made at the board’s discretion for each 50% increase in company value above a baseline. The company described the fund as a wholly owned LLC, with its Anthropic equity excluded from lobbying or political giving. Potential uses include nonprofit donations and distributions to affected populations, such as workers; these are possibilities, not confirmed payments or compensation entitlements.[24]

Anthropic said it would disclose further structure, governance and valuation-milestone details in the coming months. The announcement does not establish completed stockholder approval or a completed share transfer.[24] See October 19 news coverage.

References

  1. ↑ 1.0 1.1 1.2 1.3 Anthropic, AI Policy, current text reviewed September 12, 2026. Accessed September 12, 2026.
  2. ↑ 2.0 2.1 2.2 2.3 2.4 Anthropic, The Long-Term Benefit Trust, September 19, 2023, with subsequent trustee updates. Accessed September 12, 2026.
  3. ↑ 3.0 3.1 3.2 3.3 Anthropic, Responsible Scaling Policy Version 3.0, February 24, 2026. Accessed September 12, 2026.
  4. ↑ Business and Trade Committee, Liam Byrne to Henry de Zoete, Anthropic Model Testing, September 9, 2026.
  5. ↑ AI Security Institute, Henry de Zoete to Liam Byrne, September 15, 2026.
  6. ↑ Anthropic, PBC, Form S-1, filed October 26, 2026, preliminary-prospectus cover and “Our Governance Structure and Its Goals” summary, pp. 13–14; these complete source sections reviewed. Detailed charter, trust and risk-factor provisions require separate review.
  7. ↑ Anthropic, Anthropic confidentially submits draft S-1 to the SEC, June 1, 2026; complete announcement reviewed October 11, 2026.
  8. ↑ 8.0 8.1 Anthropic, Donating $20 Million to Public First Action, February 12, 2026. Accessed September 12, 2026.
  9. ↑ 9.0 9.1 Anthropic, Donating Another $20 Million to Public First Action, July 21, 2026. Accessed September 12, 2026.
  10. ↑ Auzinea Bacon, CNN via KEYT, “Anthropic CEO calls for ‘pacing the frontier’ of AI race amid safety concerns,” September 12, 2026.
  11. ↑ Clément Delangue, announcement of the Open Alignment Initiative, X, September 12, 2026. Public post reviewed September 12, 2026.
  12. ↑ Rishi Sunak, public post on frontier-AI oversight, X, September 12, 2026. Full public post reviewed September 12, 2026.
  13. ↑ E. Eduardo Castillo and Chan Ho-him, Associated Press, “Beijing hits back at Anthropic CEO’s call to curb China’s AI development,” September 14, 2026.
  14. ↑ New York City Council, “New York City Council to Hold Major Hearing on Artificial Intelligence Risks, Examine Need for Stronger Protections for New Yorkers,” September 16, 2026.
  15. ↑ Anthropic, “Measurements for understanding the pace of AI development inside frontier labs,” September 17, 2026.
  16. ↑ Reuters via Investing.com, “Anthropic says Claude now leads a quarter of work building its next AI models,” September 17, 2026.
  17. ↑ Buist et al. v. Anthropic PBC et al., No. 3:26-cv-10693 (N.D. Cal.), ECF 1, filed September 18, 2026, pp. 1–7, 25–29.
  18. ↑ D.C. Circuit, Anthropic PBC v. U.S. Department of War, Nos. 26-1049 and 26-1162, September 25, 2026.
  19. ↑ Josh Taylor and Tom McIlroy, The Guardian, September 28, 2026.
  20. ↑ Associated Press, September 30, 2026; FTC-spokesperson confirmation.
  21. ↑ Anthropic, “2026 Usage Policy update,” October 8, 2026; full announcement reviewed.
  22. ↑ Anthropic, “Investigating unintended model actions in our evaluations and internal use,” October 9, 2026.
  23. ↑ Anthropic, Preparing for the 2026 US midterms, October 14, 2026.
  24. ↑ 24.0 24.1 Anthropic, “Investing in the public benefit,” October 18, 2026; complete announcement reviewed October 19, 2026.
Retrieved from "https://ailawwiki.com/index.php?title=Anthropic&oldid=7482"