Anthropic pitches draft bill setting thresholds for federal AI intervention

Anthropic has privately pitched lawmakers a draft bill that would limit when federal officials could halt frontier AI models, Mashable reports, citing POLITICO's reporting and a copy of the draft. Intervention would require a foreseeable, imminent risk of more than $1 billion in damages or of killing or seriously injuring more than 50 people. The threat would also have to involve a bioweapon, a model escaping developer control or a cyberattack. The proposal would establish outside evaluators, impose duties on labs and override some state laws. Commerce would be the only agency allowed to seek a halt. Anthropic has not publicly endorsed a federal bill.