Anthropic-linked draft bill would let courts halt AI models only over imminent mass-harm risks
Anthropic official brand assets (anthropic.com)
The draft would let a D.C. court halt a frontier model only over imminent risks of 50+ casualties or $1 billion in damage, with Commerce as sole enforcer.
A draft federal AI safety bill linked to Anthropic would allow the government to seek a court order pausing a frontier AI model only when it poses an imminent risk of mass harm, according to the text of the 37-page discussion draft. Politico Pro first reported the proposal, and the document began circulating publicly on Oct. 19. The draft itself does not name Anthropic.
The bill defines catastrophic risk as death or serious injury to more than 50 people, or more than $1 billion in property damage, from a single incident. Covered risks include help building biological weapons, models evading developer control, offensive cyber operations, and automated AI research that would accelerate those threats.
Commerce would hold enforcement power
The Commerce Department would receive safety frameworks and incident reports and could ask the U.S. District Court for D.C. to suspend a model for an initial 45 days, renewable once, according to a review of the text by PPC Land. Commerce would carry the burden of proof. Large developers would need an independent evaluator, and incidents posing imminent risk of death would have to be reported within 24 hours.
The draft would bar states from regulating developer safety disclosures, third-party evaluations and incident reporting, including existing laws such as California’s SB 53. It is marked “Discussion Draft” and was last edited Sept. 29.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.

