U.S. Congress Proposes Restrictions on Government Intervention in Advanced AI Models

Anthropic is advocating for Congress to impose strict limits on federal authorities when it comes to prohibiting advanced AI models from providing services to users. According to a confidential draft bill obtained by POLITICO, government intervention would be tied to specific catastrophic threats and would require judicial approval to suspend the use of a model. The proposal aims to provide clearer legal frameworks for cutting-edge AI developers while limiting the circumstances that could trigger government action, raising questions about how much oversight regulators should impose on increasingly powerful systems.

Strict Conditions for Intervention

Under the proposal, federal authorities would only be able to seek suspension of a model when there is a foreseeable and imminent threat. Eligible risks must involve losses exceeding $1 billion or more than 50 deaths or serious injuries. The bill limits these threats to three scenarios: the creation or release of biological weapons, a model operating outside its developer's control, or cyberattacks. Other forms of harm would not fall within this specific intervention framework. Anthropic states that the legislation reflects the advanced AI framework it has been promoting among federal and state legislators since June.

“These are reasonable policies that we believe Congress should adopt,” an Anthropic spokesperson stated.

Independent evaluators will be responsible for overseeing compliance. They will register with the Department of Commerce and conduct regular assessments of cutting-edge AI developers and their models. Evaluators can be compensated by the companies they assess, provided that their pay does not depend on the results of their evaluations. If an assessment identifies an imminent threat covered by the bill, the evaluators must notify the Department of Commerce within 72 hours of completing their review.

The courts will control the suspension; the proposal would prevent the Department of Commerce from shutting down models on its own. Instead, the department would need to seek an order from a federal judge in Washington, D.C. The draft also prohibits other federal departments, agencies, and officials from suspending the development, deployment, or internal use of models. This restriction could reduce the risk of sudden enforcement actions against AI developers. This issue is particularly relevant in light of reports on U.S. export controls, which reportedly excluded two advanced Anthropic models over the summer. This incident raised concerns in the tech industry about government officials limiting access to powerful systems.

Intense Debate within Congress

However, the bill would also prioritize public risk disclosures, independent evaluations, and significant event reporting requirements in certain states. This provision could reignite debates over whether Washington should establish national standards or maintain stronger state protections.

Congress is weighing competing proposals as Anthropic's draft enters a heated legislative debate on how the U.S. should regulate advanced AI. Representatives Lori Trahan and Jay Obernolte are advancing the FRONTIER Act, which would authorize government intervention in risky models and supersede certain state laws. In the Senate, Majority Leader John Thune is working with Senators Amy Klobuchar and Ted Cruz on legislation addressing catastrophic AI risks.

However, some safety advocates and Senator Maria Cantwell are calling for a stronger federal review process. Other lawmakers are pushing for more comprehensive restrictions. Representatives Greg Casar and Senator Bernie Sanders are advocating for a pause on the development of advanced AI and a ban on superintelligence.

Anthropic's proposal takes a narrower approach. It would require cutting-edge developers to establish safety frameworks, provide transparent reporting for new model releases, protect whistleblowers, and report significant events to the Department of Commerce. Whether lawmakers accept these restrictions will help determine how much power the federal government has over advanced AI systems. This debate also focuses on whether public protection can be achieved through a single national framework without granting regulators excessive discretion.

Challenges of Government Regulation of AI

As AI technology rapidly evolves, effective regulation has become a significant issue. The bill proposed by Anthropic aims to provide a clear legal framework for developers of advanced AI while limiting arbitrary government intervention, reflecting concerns about balancing technological innovation with public safety. With differing views on regulatory measures within Congress, future legislation will influence the direction of AI development and its societal impact. This debate concerns not only the safety of technology but also how to find an appropriate boundary between fostering innovation and safeguarding public interests.