Anthropic has submitted a draft federal bill to US lawmakers concerning the safety of frontier AI models. Developed in mid-October 2026, the proposal serves as a framework for how the government should regulate the industry. It is currently a legislative pitch rather than a formal bill.
The proposal establishes specific thresholds for federal intervention. Regulators could block an AI model only if it poses a foreseeable, imminent risk of causing over $1 billion in damages or threatening the lives or safety of more than 50 people. These triggers are limited to scenarios involving biological weapons, loss of control, or cyberattacks.
Under the draft, the Department of Commerce would hold the exclusive authority to seek court orders to halt the deployment of a model. The bill also mandates independent evaluations, requiring auditors to report identified risks to the Department of Commerce within 72 hours.
The proposal incorporates elements from California’s SB 53, signed in 2025, while setting higher bars for federal action. It aligns with the company's Advanced AI Framework, which has been promoted to officials since June 2026. Anthropic has committed to refining the document's language following feedback.
Critics argue the bill is an attempt by the industry to self-regulate and address concerns previously raised by the company's own leadership. Advocates for AI safety contend that the proposed thresholds are too lenient compared to other legislative options, such as the bipartisan FRONTIER Act.
