Anthropic has circulated draft US legislation that would create a court-based process for stopping an AI model whose release poses a foreseeable and imminent catastrophic risk, according to reporting cited by TechCrunch.
The proposal defines that threshold as a risk of more than 50 deaths or over $1 billion in property damage arising from a biological weapon, a cyberattack or a model escaping its developer’s controls. Registered independent evaluators would assess systems and notify the Commerce Department within 72 hours when they find such a risk.
Commerce could not block distribution on its own. It would need to ask a federal judge in Washington, DC, for an order. Evaluators could be paid by the AI companies they examine as long as compensation is not tied to the result, an arrangement that still leaves questions about financial influence.
The text is a company-backed draft, not enacted law or an introduced congressional bill. Anthropic says it reflects a framework it has advocated since June. Its quantitative thresholds resemble California’s SB 53, but the proposed federal mechanism puts the final decision with a court rather than an agency.