Anthropic Proposes Federal Court Authority to Halt High-Risk AI Models

Anthropic reportedly presented U.S. lawmakers with an unpublished draft bill allowing courts to halt frontier AI models that pose foreseeable, imminent risks of more than $1 billion in damage or death or serious injury to more than 50 people. Covered threats include biological weapons, AI-enabled cyberattacks, and developers losing control of models or agents. Only the Department of Commerce could request a court-ordered suspension, centralizing federal authority to seek such intervention.

The proposal would impose testing, transparency, independent evaluation, and serious-incident reporting requirements on developers. Independent evaluators registered with Commerce would assess models and report imminent risks meeting the thresholds within 72 hours, triggering a federal judicial review of whether to halt release. Drawing on Anthropic’s Advanced AI Framework, the draft would make catastrophic cyber risk a potential basis for restricting model deployment, but it is not enacted law; reported White House opposition to mandatory AI regulation leaves its legislative prospects uncertain.