Anthropic quietly pitches Congress its own AI safety bill…

A confidential Anthropic draft, first reported by Politico on Oct. 16, would let the government block frontier models only for $1B+ catastrophic risks — and override state AI disclosure, evaluation, and incident-reporting laws.

Anthropic representatives have quietly circulated a confidential draft federal AI safety bill on Capitol Hill, Politico reported Oct. 16, citing three people familiar with the discussions. The draft is unusual less for what it would allow Washington to do than for what it would stop every other level of government from doing.

The threshold for federal intervention is set deliberately high. The government could block a frontier model only on a foreseeable, imminent risk exceeding $1 billion in damages or killing or seriously injuring more than 50 people, and only from a bioweapon, a cyberattack, or a model evading its developer’s controls. Politico notes the language mirrors California’s SB 53. Independent evaluators would register with Commerce, be paid by the labs regardless of findings, and have 72 hours to report any “imminent enumerated risk.” Only a federal judge in the D.C. District Court, acting on a Commerce request, could suspend a model. Every other agency is barred.

The quieter clause is the one that matters for everyone downstream of the labs. The draft would preempt state laws on public risk disclosure, third-party evaluation, and critical-incident reporting, the exact regimes already live in California, Illinois, and New York.

No lawmaker has sponsored it yet. It lands against the House FRONTIER Act and the Thune-Klobuchar-Cruz Senate negotiation, already facing pushback from Sen. Maria Cantwell, whose own blueprint demands pre-release audits. Democrats have resisted similar preemption language before.

“These are sensible policies we believe Congress should act on,” an Anthropic spokesperson told Politico.

The pitch arrives the same week as the FTC’s parallel subpoenas of OpenAI and Anthropic as eight states sued, and against the backdrop of the Trump administration’s AI liability posture leaving deployers exposed. With Congress’s AI calendar effectively dead until 2027, the preemption fight moves to a Senate Commerce hearing. A lab is now writing the ceiling on its own regulation and the floor beneath every state that tried to go first.

Anthropic quietly pitches Congress its own AI safety bill — and it would preempt state rules

Sources

Sources