Anthropic's draft AI safety bill would pause models over 50+ deaths

Politico obtained a confidential draft of Anthropic's proposed AI safety bill. It would let the government block a model's release only for a "foreseeable, imminent risk" of over $1 billion in damages or the death or serious injury of more than 50 people from a cyberattack, loss of control, or bioweapon. Only a federal judge in Washington, D.C., could suspend a model.

• Threshold: over $1 billion in damages or 50+ deaths
• Risk must involve cyberattack, loss of control, or bioweapons
• Independent evaluators report to Commerce Dept. within 72 hours
• Evaluators may be funded by the AI labs themselves