Politico reported on October 16, 2026, citing three people familiar with the effort, that Anthropic representatives have quietly pitched Capitol Hill lawmakers on a confidential draft federal bill governing frontier AI safety and security.

Under the draft, federal officials could seek to suspend an advanced AI model only if it poses a foreseeable and imminent risk of causing more than $1 billion in damages, or killing or seriously injuring more than 50 people.

Those thresholds apply only to three threat categories: creation or release of biological weapons, cyberattacks, and AI systems escaping their developers' controls. An Anthropic spokesperson confirmed the draft reflects the company's Advanced AI Framework released earlier this year.

It is one thing for a frontier lab to lobby for regulation and another to circulate legislative text a member of Congress could introduce largely as written.

Anthropic has spent 2026 arguing that AI companies cannot police themselves — its head of public policy, Sarah Heck, has said 'we can't be checking our own homework.'

Yet the draft would concentrate suspension authority in a single path: only the Commerce Department could petition a judge, and only a federal judge in the U.S. District Court for the District of Columbia could order a model halted.

Other federal departments, agencies and officers would be barred from independently suspending covered models, a design critics may read as self-interested venue selection rather than prudent centralization.

The draft would require outside evaluators registered with the Commerce Department to periodically assess AI developers and their frontier models. Companies could pay those evaluators, provided compensation was independent of the findings.

If an assessment identified an imminent enumerated risk meeting the bill's criteria, evaluators would have 72 hours after completing their review to report it to Commerce, which could then seek a court order suspending the model.

According to CNA, the Commerce Department did not immediately respond to a Reuters request for comment, and the draft has not been formally introduced in Congress.

The draft reportedly includes provisions that would override certain state AI safety laws, limiting states' ability to impose separate testing, disclosure and safety requirements — a federal preemption fight that has defined AI policy in 2026.

The stakes are personal for Anthropic too: it was banned from the Pentagon in February and was forced in June to suspend its most advanced models for nearly three weeks, and a federal appeals court upheld that 'supply chain risk' designation, per Law Commentary.

CEO Dario Amodei has called on the industry to slow development, while President Trump has dismissed such calls as 'negative forces.'

Anthropic's effort lands alongside competing proposals. According to Law Commentary, Democratic Sens.

Mark Warner, Brian Schatz and Andy Kim introduced the Artificial Intelligence Risk Management and Security Act of 2026 (S. 5576) on September 29, which would create a federal AI Safety Board and enforceable safety requirements; Sen.

Ted Cruz blocked its unanimous-consent passage and it remains in committee. Charlie Bullock of the Institute for Law & AI called Anthropic's draft 'pretty reasonable' but 'weaker' than the House's FRONTIER Act, per CNA.

Anthropic separately granted six METR researchers employee-like access to its internal systems on October 15, Law Commentary reports.

The privately circulated draft has yet to be formally introduced, and any bill would need approval from both chambers and the president's signature to become law. Congress is out of session until after next month's midterm elections, leaving time for lobbying on all sides.

Its proposed limits on federal intervention and state regulatory authority would face scrutiny from lawmakers who want multiple checks on AI companies.

The company's push for third-party evaluation — including the METR arrangement and work with SecureBio and Accenture — signals that independent oversight will remain central to whatever framework emerges.