Anthropic's draft AI bill sets a $1B or 50-person risk bar for model shutdowns
Politico reports that Anthropic's proposal would channel model suspension through Commerce and a federal judge, while overriding some state safety rules.
Anthropic is pitching lawmakers a federal AI safety bill that would sharply define when the government could stop a frontier model, according to an October 16 Politico exclusive. The outlet says it obtained a confidential draft and spoke with three people familiar with the effort. An Anthropic spokesperson said the bill represents proposals in the company’s Advanced AI Framework, which it has advocated since June. This is a reported draft proposal, not an enacted law.
A risk threshold, not a death counter
Politico reports that the draft would permit blocking a model’s release or use only for a foreseeable, imminent risk of more than $1 billion in damages, or killing or seriously injuring more than 50 people. The risk would have to arise from a biological weapon, a cyberattack, or a model evading its developer’s controls.
The distinction matters: the reported test concerns a prospective risk. It does not require those deaths or losses to happen first, and the 50-person threshold includes serious injuries. Nor does this reported suspension provision establish that smaller harms are exempt from every other law.
Commerce would go to court
Independent evaluators registered with the Commerce Department would periodically assess frontier labs and models, Politico says. Labs could pay them provided payment was not contingent on evaluation results. An evaluator finding an imminent enumerated risk would have to report it to Commerce within 72 hours of completing its assessment.
Commerce could then ask a federal judge in the District of Columbia to suspend a model. According to the report, the draft explicitly bars other US departments, agencies or officers from suspending its development, deployment or internal use. That concentrates the proposed intervention route rather than giving Commerce an unrestricted unilateral shutdown power.
The draft would also preempt some state rules covering public risk disclosure, third-party evaluation and critical-incident reporting. Separately, it would require safety frameworks, transparency reports for new frontier models, whistleblower protections and critical-incident reports to Commerce. These provisions make the proposal more than a shutdown threshold: it would reshape who oversees frontier labs and which state requirements survive. Their eventual scope depends on any text lawmakers actually introduce and pass.