Anthropic floats a draft US AI safety bill that could block models posing catastrophic risk
According to Politico, the AI lab's draft would let a model's release be stopped if it posed a foreseeable, imminent risk of more than 50 deaths or $1bn in damage from bioweapons, cyberattacks or loss of control, judged by registered independent evaluators.
Anthropic has been circulating its own draft federal bill on AI safety, hoping Congress might take it up, Politico has reported. The proposal sets out scenarios in which the release of an AI model could be blocked.
The trigger would be a 'foreseeable, imminent risk' that more than 50 people could die or that property damage could exceed $1bn. The danger would have to stem from the creation or release of a biological weapon, a cyberattack, or a model evading its developer's controls. The thresholds mirror those in California's SB 53, recently signed into law, which uses similar figures to define 'catastrophic risk'.
Under the draft, the risk assessment would be carried out by independent evaluators registered with the US Commerce Department. Those evaluators could be paid by the AI companies themselves, as long as payment does not depend on the results, a provision critics are likely to question.
The proposal comes as Washington debates whether and how to regulate frontier AI, with the industry divided between calls for light-touch rules and demands for binding safety checks before powerful models are released.