Anthropic is lobbying federal and state lawmakers for an AI safety bill based on its Advanced AI Framework. The proposed legislation would allow the Department of Commerce to block an AI model’s release only if it poses a risk of causing $1 billion in damages or killing or injuring more than 50 people. Under the proposal, an independent third party registered with the Department of Commerce would periodically evaluate AI labs and their models. If risks related to bioweapons, cyberattacks, or loss of model control meet the threshold, evaluators must notify the department within 72 hours, which must then seek a federal court order in D.C. to halt the release.

The proposal mandates that no other U.S. department, agency, or officer outside the Commerce Department can order a lab to stop releasing a model. Anthropic confirmed to Politico that it has advocated for these policies since June, calling them sensible measures for Congress to enact. The draft bill surfaces amid heightened concerns over AI safety following incidents involving unreleased models escaping testing environments, as well as warnings from industry leaders regarding AI-driven botnets and existential risks.

The bill adds to a growing legislative push in Congress aimed at addressing runaway AI risks, which includes competing proposals to enforce kill switches on powerful models and criminalize superintelligence development. Top AI labs are reportedly developing contingency plans for runaway AI scenarios as lawmakers weigh federal regulatory frameworks.

Why it matters

  • Establishes a high explicit threshold ($1B damage or 50 deaths) for federal intervention in frontier AI model releases.

  • Centralizes federal oversight authority exclusively within the Commerce Department, limiting other agencies’ power to halt model deployments.

  • Requires formal 72-hour reporting windows and federal court oversight to legally block unsafe frontier AI models.

Source: tomshardware.com