Anthropic Drafts Bill to Block Dangerous AI Models

Anthropic has been circulating a draft AI safety bill it hopes Congress will adopt. The legislation would allow an AI model's release to be blocked if independent evaluators determine it poses a foreseeable, imminent risk of killing over 50 people, causing over a billion dollars in property damage, or enabling biological weapons, cyberattacks, or a model evading developer controls. Evaluators would register with the Commerce Department and report risks within 72 hours. A Washington judge would then decide on halting a release. Federal agencies could not block models independently.