Anthropic is privately pitching Congress its own draft federal AI safety bill, first reported by Politico on Oct 16 and confirmed by Reuters. The draft would let the government block a frontier model only for foreseeable, imminent catastrophic risk: over $1 billion in damage or killing or seriously injuring more than 50 people, per BullTheoryio and WhaleInsider.

Anthropic is privately pitching Congress its own draft federal AI safety bill, first reported by Politico on Oct 16 and confirmed by Reuters. The draft would let the government block a frontier model only for foreseeable, imminent catastrophic risk: over $1 billion in damage or killing or seriously injuring more than 50 people, per BullTheoryio and WhaleInsider.

The risk would have to come from bioweapons, a model escaping developer control, or cyberattack. Independent evaluators registered with Commerce would have 72 hours to report such risk, then only Commerce could ask a federal judge in D.C. to suspend the model, and no other agency could act. A court could not force a lab to hand over or delete weights.

Labs could pay their own evaluators as long as pay is not tied to the verdict, with evaluators shielded from liability. Anthropic told Politico this reflects proposals in its Advanced AI Framework advocated since June.

White House tech co-chair David Sacks criticized the draft as catching too many companies by setting the frontier around 10^25 compute. Anthropic has not posted the draft on its own X account, and the full Politico/Reuters texts were not directly readable here, so details above rely on their syndication in posts and Telegram reports like markettwits.