On October 16, 2026, Politico reporter Owen Dahlkamp reported that Anthropic representatives had pitched lawmakers on a confidential draft federal AI safety bill. Politico said it had obtained the draft and cited three people familiar with the effort. The proposal would tie federal action to block a model’s release or use to a foreseeable, imminent risk of more than $1 billion in damages or more than 50 people killed or seriously injured.
Which risks would qualify?
The reported draft names three risk categories: creating or releasing a biological weapon, a model evading its developer’s controls, or a cyberattack. The proposed intervention threshold would apply to foreseeable, imminent risks involving those categories.
Who could seek to halt a model?
Tom’s Hardware’s account of the proposal describes an independent evaluator registered with the U.S. Department of Commerce, which would periodically assess the lab and its frontier models. An evaluator identifying a qualifying imminent risk would have 72 hours to notify Commerce. The lab could pay the evaluator if the payment did not affect its independence, according to Tom’s Hardware.
Commerce would have the sole department-level role in seeking a halt. After Commerce presented evidence, a federal judge in the U.S. District Court for the District of Columbia would decide whether to stop the model’s release.
The use of outside evaluators also places the proposal in a wider discussion of oversight, explored in NeoTeo’s earlier coverage of external AI oversight.
How the proposal differs from Anthropic’s public framework
Anthropic’s June 2026 Advanced AI Framework recommends against federal preemption of state law unless a rigorous federal regime meets or exceeds the framework’s strongest measures. It also recommends limiting preemption to specified frontier-governance functions, leaving states authority outside those areas. Politico separately reported that the confidential draft would preempt some state AI safety laws.
The reporting deadlines concern different duties. The draft’s reported 72-hour period is for an evaluator to notify Commerce of a qualifying imminent risk. Under the public framework, a covered developer would report a defined Critical Safety Incident to the designated agency within 15 days of discovery.
A threshold for government intervention is separate from a rule about civil liability. Anthropic’s public framework says federal compliance should not itself confer immunity, a safe harbor, or a presumption against liability under otherwise applicable state law.
