Anthropic proposes AI safety bill with potential model kill switch

• Anthropic has proposed draft legislation for AI safety.
• The bill defines catastrophic risk as over 50 potential deaths or over $1 billion in property damage.
• Independent evaluators would assess AI models for potential risks.
• The Commerce Department would have 72 hours to act on evaluator reports.
• Federal agencies would be prohibited from independently halting model distribution.

AI company Anthropic has proposed draft legislation aimed at enhancing AI safety, which it has been advocating for with lawmakers since June. The bill outlines scenarios where the distribution of an AI model could be blocked if it presents a "foreseeable, imminent risk." This risk is defined by potential outcomes such as over 50 people dying or more than $1 billion in property damage resulting from issues like a biological weapon creation, evasion of developer controls, or a cyberattack. The proposed framework involves independent evaluators, registered with the U.S. Commerce Department, who would assess models for catastrophic risks. If such risks are identified, evaluators would have 72 hours to notify the Commerce Department, which would then need to seek a judicial order to halt the model's release, explicitly forbidding federal agencies from acting unilaterally.