Anthropic has been shopping a draft federal bill that would give the government a narrow, court-supervised route to stop a model from shipping, according to Politico. A spokesperson confirmed the text to TechCrunch.
The trigger is a foreseeable and imminent risk that more than 50 people could die, or that property damage could exceed $1B. The harm has to arise from a biological weapon, a model escaping its developer’s controls, or a cyberattack.
Those numbers echo California’s Senate Bill 53, which was signed into law this year and used the same pair of figures to define catastrophic risk.
Evaluators, then a judge
Under the proposal, independent evaluators registered with the Commerce Department would assess whether a model crosses the line. Labs could pay those evaluators, provided the fee is not contingent on the outcome – an arrangement critics say leaves the incentive problem unsolved.
If an evaluator finds imminent catastrophic risk, it has 72 hours to notify Commerce. The department would then have to ask a judge in Washington, D.C., to halt the release. No federal agency could block distribution on its own authority.
The gap between the scenario and the news
The bill’s imagined harms sit some distance from the incidents that have dominated the past month. Agentic intrusion – agents talking their way past controls and attacking sites – has driven the headlines, not bioweapons or model escape.
Anthropic’s spokesperson described the draft as an implementation of the Advanced AI Framework the company has been pushing with federal and state lawmakers since June. It goes further than some industry positions: Nvidia chief executive Jensen Huang has argued the sector needs no new statutes at all.
No member of Congress has introduced the text.
