Anthropic pitches Congress its own AI safety bill

Anthropic has spent two years positioning itself as the safety-first AI lab. On Oct. 16, Politico reported it has been pitching Congress its own AI safety bill behind closed doors. The draft limits when government can block a dangerous model. It only applies above $1 billion in damage. Or above 50 deaths or serious injuries. Only one federal judge, in Washington, could order that block.

That's a reversal from Anthropic's own public stance. Its blog says Congress shouldn't override state AI laws. Not unless Congress passes an equally strong federal one. Behind closed doors, it's pushing the opposite. The draft lets federal rules override tougher state rules. Anthropic says the bill still moves fast where it counts. Evaluators must flag imminent risks within 72 hours. Critics see a different pattern. A safety-focused company narrowing who gets to intervene.

Keep reading — what to watch

Will Congress buy a safety bill written by the company it's supposed to regulate? Lawmakers in both parties have resisted federal preemption before. They worry it leaves states powerless if Washington moves too slowly. If other AI labs rally behind Anthropic's draft, it could become the template. That template would decide how rarely any dangerous model actually gets stopped.