Anthropic Wrote Its Own AI Safety Bill. Here's What It Allows

Jeff's Daily Pick. Anthropic quietly handed lawmakers its own AI safety bill. Washington could halt a model only for risks above $1 billion or 50 casualties, and only through one federal judge.

In the United States, representatives of Anthropic, the developer of Claude, have pitched Capitol Hill lawmakers on a draft bill to regulate frontier AI safety and security, Politico reported on October 16, 2026, citing three people with knowledge of the effort and a confidential draft it obtained. Reuters relayed the report the same day. An Anthropic spokesperson said the bill "represents the proposals outlined in Anthropic's Advanced AI Framework, which we have been advocating for with federal and state lawmakers since June." The draft itself has not been published.

When a model could be stopped

According to the reports, the government could block a model's release or use only if it poses a foreseeable, imminent risk of more than $1 billion in damages, or of killing or seriously injuring more than 50 people. The risk must come from one of three sources: the creation or release of a biological weapon, the model evading its developer's controls, or a cyberattack. Politico noted that the wording mirrors California's AI laws, including SB 53, signed in 2025.

Who decides

Independent evaluators registered with the Commerce Department would periodically assess each lab and its frontier models. Labs could pay them, provided payment is not tied to the outcome. If evaluators find an "imminent enumerated risk," they must report it to Commerce within 72 hours of completing their assessment. Commerce could then ask a federal judge of the U.S. District Court for the District of Columbia to suspend the model. The draft explicitly bars any other federal department, agency or officer from suspending a model's development, deployment or internal use, and Politico reported it would preempt some state AI safety laws. The Commerce Department did not immediately respond to Reuters.

The framework behind it

Anthropic's Advanced AI Framework, published in June 2026, says governments should have "a way to block or deter deployment of models that pose significant catastrophic risks," and lays out several options, from an agency imposing fines or deployment bans to a lighter model where the agency must sue in court. The draft as reported takes the court route. The framework also sets its own conditions on preemption: Congress should displace state law only if it enacts a federal regime that "meets or exceeds the strongest measures proposed in this framework," and then only narrowly. It would cover developers that train models with more than 10²⁵ operations, or earning more than $500 million a year from AI, or spending more than $1 billion a year on AI research.

The draft arrives as Anthropic prepares for a possible public listing, covered in our piece on its $16.8 billion quarter. A company proposing the rules for its own product is not new in Washington. What matters for readers is the trigger: below $1 billion or 50 casualties, the draft as described gives the government no power to halt a model. For a different proposal, see our piece on 400 scientists asking for an AI referee.