Anthropic pitches AI safety bill with a high bar for intervention

Anthropic, the maker of Claude, has circulated a draft federal bill that would govern only the most capable AI models, rather than everyday chatbots. It would require leading developers to test their systems, disclose safety findings, submit to independent evaluations, and maintain strong security—creating national oversight for models that could help create biological weapons, carry out cyberattacks, or escape their developers’ controls. But it also sets a narrow gate for intervention: the Commerce Department could ask a judge to halt a model only when it poses a foreseeable, imminent risk of more than $1 billion in damage or more than 50 deaths or serious injuries. The proposal would also override some state rules, making it a bid by a leading AI company to define both its industry’s obligations and the limits of federal power before Congress has passed a broad AI law.