- Anthropic proposed a federal AI safety bill to members of the U.S. Congress, Politico reported on October 17, 2026.
- The proposal would restrict when the government may halt an advanced AI model while requiring independent evaluations, risk reporting and developer safeguards.
- The draft would also roll back some state-level requirements, an issue that remains contested in Congress.
Anthropic has proposed federal legislation governing advanced artificial intelligence models, according to an October 17, 2026, report by Politico. The draft would allow government intervention only when a model presents specified severe risks and would limit the authority of federal agencies to suspend its development or use.
Thresholds for halting an AI model
According to Politico, Anthropic delivered the draft bill to members of Congress. The proposal seeks to establish safety rules for advanced AI models while restricting the government’s power to intervene in their development and deployment.
Under the draft, the government could seek to block the release or use of a model only if it poses a foreseeable and imminent risk of more than $1 billion in damage or the death or serious injury of more than 50 people. The covered threats include the creation or release of biological weapons, a model bypassing a developer’s safeguards, or a cyberattack.
The Department of Commerce could petition a court to halt a model’s release. The bill would prohibit other federal agencies and officials from independently suspending the development, deployment or internal use of covered systems. A federal court in Washington would make the final decision on whether to halt a model.
Audits, safeguards and incident reporting
The proposal would require developers of advanced models to build and maintain safety systems, publish transparency reports for new models, protect whistleblowers who report violations and notify the Department of Commerce about critical safety incidents.
Independent evaluators registered with the department would periodically audit developers and their models. Companies could pay for the evaluations provided that compensation is not tied to the results. If an assessment identifies an imminent risk as defined by the bill, the evaluator would have to notify the department within 72 hours after completing the review.
An Anthropic spokesperson said the document is based on the company’s advanced AI safety framework.
“This bill represents the proposals outlined in Anthropic’s Advanced AI Framework, which we have been advocating for with federal and state lawmakers since June. These are sensible policies we believe Congress should act on,” the spokesperson said.
State rules and other congressional proposals
The draft would roll back some state-level requirements covering public disclosure of risks, independent model evaluations and critical-incident reporting. Some Democrats fear that a single federal standard would weaken rules already adopted in California, Illinois and New York. Technology industry representatives argue that differing state requirements make compliance more difficult.
Other proposals are under discussion in Congress. In the House of Representatives, Reps. Lori Trahan and Jay Obernolte are advancing the FRONTIER Act, which would also permit dangerous model deployments to be paused and would preempt some state laws. In the Senate, Majority Leader John Thune and Sens. Amy Klobuchar and Ted Cruz are negotiating legislation intended to address catastrophic AI risks.
Some politicians and technology-safety organizations are seeking stronger government oversight, while some progressive lawmakers are calling for a pause in advanced AI development.
Anthropic’s dispute with the Trump administration
Anthropic’s conflict with President Donald Trump’s administration escalated in early 2026 over disagreements about military uses of AI. Authorities subsequently placed the company on a supply-chain risk list and restricted federal agencies from working with it.
The U.S. government later demanded restrictions on foreign access to the Claude Fable 5 and Mythos 5 models, prompting Anthropic to suspend support for both models for all customers. Federal Judge Rita Lin ruled that adding Anthropic to the blacklist was unlawful, finding that the authorities had not demonstrated a genuine national-security threat and that the restrictions violated the First and Fifth Amendments to the U.S. Constitution.
Anthropic also faced broad export restrictions over the summer that cut off access to two of its most advanced models. The administration’s decision raised concerns in the technology sector about officials’ discretion to restrict access to AI.
The Trump administration separately asked OpenAI to limit the initial rollout of GPT-5.6 to a small group of government-approved partners so the model could be tested on national-security grounds before a public release.
Source: Incrypted
