Anthropic backs AI safety bill that halts model releases at 50 deaths or $1B in damages

Anthropic is lobbying for a bill letting the US Commerce Department block an AI model's release if it risks killing more than 50 people or causing over $1 billion in damages. An independent evaluator would assess models, and a federal judge in DC would decide whether to halt the release.

• Threshold: more than 50 deaths or $1 billion in damages
• Evaluator must report to Commerce within 72 hours
• A DC federal judge decides whether to halt the release
• Only Commerce could order a lab to stop releasing a model