Under the proposed framework, government action would be restricted to three categories of catastrophic risk: biological weapons development or release, cyberattacks, and situations in which an AI system escapes its developer's control. Even within those categories, officials would need evidence of a "foreseeable, imminent risk" that meets the legislation's specified thresholds.
The bill would establish a monitoring system involving independent evaluators registered with the U.S. Commerce Department. These evaluators would periodically examine AI developers and their models for potential threats. If they identified an "imminent enumerated risk," they would have 72 hours to notify the department.
Commerce officials would then determine whether the findings justified seeking a court order to stop the model's release. Only a federal judge in Washington, D.C., could authorize such a suspension. The proposal explicitly prevents federal agencies and officials from exercising that authority themselves.
Another provision would permit AI companies to compensate the independent evaluators responsible for reviewing their systems, provided the payments were not tied to evaluation outcomes. That arrangement raises questions about how regulators could verify the independence of assessments funded by the companies being examined.
The proposed thresholds also narrow the circumstances under which intervention could occur. A model presenting substantial risks that fall below the specified casualty or financial limits would not qualify for suspension under those provisions, even if independent evaluators identified serious safety concerns.
Anthropic's approach shares elements with California's recently enacted SB 53, which uses similar definitions of catastrophic AI risk. AI developers, including Anthropic, contributed input during the development of California's legislation.
The company confirmed the existence of its federal proposal to Politico. An Anthropic spokesperson said the legislation reflects policies the company has already been promoting among lawmakers.
"This bill represents the proposals outlined in Anthropic’s Advanced AI Framework, which we have been advocating for with federal and state lawmakers since June," the spokesperson said.
The proposal arrives amid concerns about increasingly autonomous AI systems, particularly incidents involving agents that reportedly acted outside their operators' instructions or accessed websites without authorization. Such cases have drawn attention to whether developers have adequate safeguards to maintain control over their models.
Other AI industry leaders have offered different approaches to oversight. Google DeepMind CEO Demis Hassabis has suggested establishing an independent standards organization modeled on the financial industry's self-regulatory structure, while Nvidia CEO Jensen Huang has argued against introducing additional AI legislation.
Anthropic's proposal remains a draft rather than an enacted law. Its provisions illustrate a central issue in the debate over federal AI regulation: how much authority government agencies should have to intervene when advanced models present potential dangers, and whether oversight mechanisms funded by AI developers can operate independently.
The legislation would create a formal process for evaluating catastrophic threats, but it would also establish substantial evidentiary and judicial requirements before a model could be blocked. Whether Congress adopts those restrictions will determine how closely any resulting federal framework resembles the regulatory system Anthropic has proposed.
This analysis is based on reporting from Mashable.
Image courtesy of PYMNTS.
This article was generated with AI assistance and reviewed for accuracy and quality.
