What happened
On 16 October 2026, Politico reported that Anthropic representatives privately pitched a draft bill to regulate frontier AI safety and security to lawmakers. The draft would let the government block a model's release or use only if it poses a foreseeable, imminent risk of more than $1 billion in damages or of killing/seriously injuring more than 50 people, in three enumerated risk categories (creation/release of a biological weapon, a model evading its own developer's controls, or a large-scale cyberattack). Enforcement would be channelled exclusively through the Commerce Department, with independent evaluators registered with Commerce and obliged to report an 'imminent enumerated risk' within 72 hours; any halt would require Commerce to seek a federal court order. It would also bar other federal agencies (including the FTC, DOJ, Treasury and HHS) from independently suspending a model and would preempt some state AI safety laws.
Why it matters
A frontier lab is drafting the rules that would govern it, establishing thresholds for federal AI enforcement and consolidating oversight into a single agency. It would materially raise the bar for government intervention in model release, preempt state AI safety statutes, and limit other federal regulators' ability to pause AI systems — directly shaping the federal-vs-state preemption fight now underway. It has not been formally introduced.
Action needed
Track whether the bill is formally introduced; assess its implications for state AI-law compliance, federal enforcement exposure, and the broader federal preemption debate.