Anthropic's draft AI safety bill would let Commerce seek court-ordered halts of risky models

Anthropic has presented draft federal legislation to US lawmakers that would let authorities seek suspension of a frontier model when it crosses defined catastrophic-risk thresholds. According to Politico, which first reported the draft, the triggers cover a biological weapon, an AI system acting beyond human control, or a cyberattack that could imminently kill or injure more than 50 people or cause more than $1 billion in damages. Enforcement would run through the Commerce Department. The proposal follows the framework Anthropic released earlier this year, and it also addresses frontier-model testing, disclosures and independent evaluations. Commentators have questioned how high the threshold sits and whether evaluators would stay independent of the labs they assess. For founders, the bill would define when a model can be pulled from release. Even as a draft, the triggers could shape how labs scope evaluations, incident reporting and launch gates before any federal rule exists.