Anthropic pitches an AI safety bill draft to US Congress

According to POLITICO, three sources say Anthropic representatives have quietly pitched a bill draft to US lawmakers, aiming to set safety and security standards for frontier AI. The move shows the lab is trying to shape the rules for its own technology through actual legislation.

The draft raises the bar for government intervention: officials can only halt a model's release if it may cause over $1 billion in economic losses, or over 50 deaths or serious injuries, and the risk must come from bioweapons, loss of control, or major cyberattacks. The wording echoes California's SB 53 law.

The bill would also require frontier AI labs to build safety frameworks, publish transparency reports, protect whistleblowers, and report major incidents to the US Commerce Department. It would override some state AI laws, addressing industry concerns about fragmented rules and high compliance costs.