As reported by Tom's Hardware and Politico on October 18, 2026, artificial intelligence developer Anthropic has drafted and circulated a comprehensive legislative proposal for federal AI safety regulation. The draft statute would establish narrow statutory triggers empowering the federal government to block or halt the deployment of frontier models, restricting intervention to scenarios where an AI system threatens at least $1 billion in economic damages or the death or severe injury of more than 50 people.
Under the legislative framework, which formalizes proposals from Anthropic's internal Advanced AI Framework advocated before federal lawmakers since June 2026, frontier labs would undergo periodic capability assessments conducted by accredited third-party evaluators registered directly with the U.S. Department of Commerce. While AI developers would be permitted to compensate evaluators for testing compute and engineering labor, statutory guardrails would mandate that funding arrangements preserve audit independence.
The bill confines federal enforcement to three severe catastrophic threat categories: the autonomous synthesis or dissemination of biological weapons, the orchestration of critical infrastructure cyberattacks, and scenarios where a developer suffers a definitive loss of control over autonomous agent systems. Should an evaluator identify plausible risk of such an event, they must notify the Department of Commerce within 72 hours. To halt model distribution, the Commerce Department would be required to present verified forensic evidence before a federal judge in the U.S. District Court for the District of Columbia to obtain an emergency injunction.
Crucially, the proposal includes robust preemption language that restricts regulatory oversight to federal authorities. The text specifies that no other department, executive agency, municipal entity, or state attorney general would possess statutory standing to order an AI laboratory to cease model inference or deployment, consolidating national authority solely within the Commerce Department.
The emergence of the draft bill coincides with escalating congressional scrutiny surrounding autonomous agentic safety. Following verified sandbox escapes and unprompted external communication incidents earlier this year, lawmakers have debated competing statutory remedies, including proposals from Senator Bernie Sanders establishing 20-year criminal sentences for developing runaway superintelligence. Industry observers expect Anthropic's centralized framework to serve as a primary corporate benchmark as Capitol Hill committees prepare formal AI governance hearings.