News follow-up · Governance

AI self-regulation: what happened since

In 2024 the argument was whether companies could be trusted to police their own frontier models. Since then one state has made the companies' own safety frameworks a legal duty, the federal government has set out to override state laws, and the frameworks themselves have been rewritten many times.

Published 29 October 2026 · Every number links to its source and was read on 29 October 2026.

What this site reported

On 26 May 2024 the earlier version of this site reported an argument by two former OpenAI board members that self-governance by AI companies could not hold against commercial pressure and that regulation was needed.

What the record shows now

A state law. On 29 September 2025 California's governor signed Senate Bill 53, the Transparency in Frontier Artificial Intelligence Act. The governor's office describes four parts: large frontier developers must publish on their websites a framework describing how they have incorporated national and international standards and industry best practice; frontier companies and the public get a channel to report potential critical safety incidents to the state's Office of Emergency Services; whistleblowers who disclose significant health and safety risks from frontier models are protected, with a civil penalty for non-compliance enforceable by the Attorney General; and a consortium called CalCompute is to design a public computing cluster.[1] The announcement calls it "first-in-the-nation frontier AI safety legislation".[1]

A federal counter-move. Executive Order 14365 of 11 December 2025 states a policy of "a minimally burdensome national policy framework for AI" and argues that state-by-state regulation creates "a patchwork of 50 different regulatory regimes".[2] It directs the Attorney General to set up, within 30 days, an AI Litigation Task Force "whose sole responsibility shall be to challenge State AI laws inconsistent with" that policy; directs the Secretary of Commerce to publish, within 90 days, an evaluation of state AI laws identifying "onerous" ones; and makes states with such laws ineligible for certain broadband funds.[2] The order names a Colorado law as an example; it does not name California.[2]

The federal testing body. NIST's programme page now carries the name Center for Advancing Innovation and Standards for Super Intelligence, CAISSI. Its listed tasks include voluntary agreements with private developers and evaluators, unclassified evaluations of capabilities that may pose risks to national security, and representing United States interests abroad "to guard against burdensome and unnecessary regulation of American technologies by foreign governments".[3] The four news items shown on the page are three assessments of models from Chinese developers, dated July and September 2026, and a note of 16 October 2026 on incident readiness in training and evaluation.[3]

The companies' frameworks. One published framework shows how often such documents change. Anthropic's page lists nine versions of its Responsible Scaling Policy between September 2023 and July 2026, five of them in 2026, and two redacted risk reports, for February and August 2026.[4]

What it means for public bodies

The 2024 question has an answer of sorts: in California, publishing a safety framework is now a legal duty for large frontier developers, with incident reporting and whistleblower protection behind it.[1] Whether that duty survives the federal task force is the open question, and the executive order gives the Attorney General the job of testing it in court.[2] A public body buying a frontier model can read the developer's framework, but the version it read may not be the version in force a month later.[4]

See also: Frontier model safety: what happened since and AI in the public sector.

Sources

1. Office of the Governor of California, Governor Newsom signs SB 53, advancing California's world-leading artificial intelligence industry, 29 September 2025. Read 29 October 2026.
2. Federal Register (govinfo), Executive Order 14365, Ensuring a National Policy Framework for Artificial Intelligence, signed 11 December 2025, published 16 December 2025. Read 29 October 2026.
3. National Institute of Standards and Technology, Center for Advancing Innovation and Standards for Super Intelligence (CAISSI), programme page with news items to 16 October 2026. Read 29 October 2026.
4. Anthropic, Responsible Scaling Policy, page listing versions 1.0 to 3.4 and the risk reports, last updated 14 August 2026. Read 29 October 2026.

How this page was checked: the four documents were read in full on 29 October 2026; the version table was copied from the dates on the company's own page. The text of SB 53 itself was not consulted; the record used for the description of the law is the governor's announcement, read on 29 October 2026.