Anthropic Proposes Federal Standards for Blocking High-Risk AI Models

Anthropic has circulated draft federal legislation that would establish conditions for blocking the release of an advanced AI model, according to reports citing a copy obtained by Politico.

The proposal defines a “catastrophic risk” as a foreseeable, material threat that a model could contribute to the death or serious injury of more than 50 people or cause more than $1 billion in property damage in a single incident. Covered threats would include the creation or release of a biological weapon, a cyberattack, or a model evading its developer’s controls. Anthropic says the thresholds were drawn from standards in California law.

Under the draft, independent evaluators registered with the Commerce Department could assess models, including through evaluations paid for by developers as long as payment was not tied to the result. If evaluators found an imminent catastrophic risk, they would have 72 hours to notify the department. Federal officials would then need to seek an order from a judge in Washington, D.C., rather than halt distribution directly. The bill would also preempt tougher state requirements, according to Novara Media.

AI safety advocates cited by Novara Media criticized the intervention threshold and limits on government authority. Anthropic federal affairs chief Anthony Cimino said the proposal would not eliminate corporate liability or other government powers, while acknowledging that the thresholds may need reconsideration. Anthropic told Politico that the draft reflects a framework it has promoted to federal and state lawmakers since June.