Anthropic Proposes Federal AI Safety Bill Limiting Government Intervention
The privately proposed legislation would restrict federal power to block advanced AI models and preempt certain state rules. It sets thresholds for intervention, gives the Commerce Department sole authority to seek court-ordered suspensions, and requires independent evaluations and developer safety reporting.
According to POLITICO, Anthropic has privately put forward federal AI safety legislation that would restrict when the U.S. government could block advanced AI models and supersede certain state rules. Under the proposal, the government could prevent a model’s release or use only in cases involving foreseeable, imminent risks that could cause more than 1 billion in damages or threaten death or serious injury to more than 50 people. The risks would need to involve biological weapons, cyberattacks or models escaping developer controls.
Only the Commerce Department could ask a court to suspend a model. Other federal agencies could not independently stop its development, deployment or internal use. Registered evaluators would review frontier models and report qualifying imminent risks within 72 hours. Labs could pay the evaluators, but compensation could not be tied to their findings.
Developers would be required to establish safety frameworks, issue transparency reports, protect whistleblowers and report critical incidents. The proposal would supersede certain state AI disclosure, evaluation and reporting laws, despite Anthropic’s previous support for state-level The proposed safety legislation comes after U.S. export restrictions shut down two advanced Anthropic models. Congress remains divided over the extent of federal oversight powers and protections for state laws.