According to Politico, Anthropic is privately lobbying U.S. Congress on a draft frontier AI safety bill that would set the threshold for government intervention in AI models at: when a model faces foreseeable and imminent specific risks capable of causing over $10 billion in damages or resulting in more than 50 deaths or serious injuries. The draft restricts regulatory triggers to three categories of risk: models being used for biological weapons development or proliferation, models escaping developer control, and models triggering cyberattacks.

The draft also mandates AI developers conduct transparent testing and undergo independent third-party assessments; upon identifying relevant risks, assessment bodies must report to the U.S. Department of Commerce within 72 hours, which would then petition a federal court to suspend model operations, with final decisions resting with the judiciary. The proposal additionally prohibits other U.S. federal agencies and officials from independently halting the development, deployment, or internal use of AI models. This initiative originates from Anthropic’s advanced AI framework released earlier this year and remains in the lobbying phase, not yet law; if enacted, it could constrain the U.S. government’s ability to take emergency interventions against frontier AI models.