Overview
- Politico reported and Anthropic confirmed on Oct. 16 that the company quietly circulated a confidential draft federal bill that would govern 'frontier' AI safety and security.
- The draft would let the government seek a court-ordered suspension only for narrowly defined imminent threats that cause over $1 billion in damages or more than 50 deaths or serious injuries tied to biological weapons, models escaping control, or cyberattacks.
- It would require periodic independent evaluations registered with the Commerce Department, let companies pay evaluators so long as pay is not outcome‑contingent, and force evaluators to report imminent listed risks to Commerce within 72 hours.
- The proposal channels enforcement through Commerce asking a District of Columbia federal judge to halt models, bars other agencies from independent suspensions, and would pre-empt some state laws including parts of California’s rules.
- Critics say the static 10^25 FLOPs plus financial test and the high damage/death thresholds would rapidly capture many firms while leaving most real harms outside federal reach, and the draft has not been introduced in Congress.