Anthropic Proposes Federal Bill for Artificial Intelligence Models

Claude AI model developer Anthropic has drafted a bill that ties federal government intervention in the development and deployment of frontier AI models to specific criteria. According to the draft, the federal government would be able to intervene if a model carries a foreseeable and imminent risk of causing more than 1 billion dollars in financial damage or more than 50 deaths or serious injuries.

While the mechanism introduced under the bill involves the Department of Commerce, the authority to issue a suspension order is envisioned to rest with a federal judge in the U.S. District Court for the District of Columbia. The draft does not permit any other federal department or agency to use this mechanism. Additionally, it proposes an independent auditor system to evaluate safety practices, requiring a report to be submitted to the relevant department within 72 hours if a risk is detected.

Critics such as Sneha Revanur from the AI safety group Encode state that the thresholds in the bill are too high and leave no lower-tier intervention space for smaller-scale cumulative risks. Considering that, according to IBM's 2026 Cost of a Data Breach Report, AI-driven breaches average 6 million dollars, even the cumulative damage of thousands of incidents might not reach the single-incident threshold of 1 billion dollars. The draft does not clearly explain how judges and auditors will calculate these cumulative losses.