Highlight any text to annotate

This bill establishes a federal program called the Advanced Artificial Intelligence Nuclear Evaluation Program, overseen by the Department of Energy, to test advanced AI systems and assess their potential to cause nuclear security incidents. Large AI developers—defined as companies that have spent at least $2 billion on AI development over five years—are required to participate in the program, submit their AI systems for testing, and provide access to model weights and safety-related documentation. The program will conduct security testing including "red-teaming" exercises simulating attacks by sophisticated adversaries, identify vulnerabilities and risks, and provide participants with detailed reports on findings. Developers who refuse to participate or submit false information face civil penalties up to $1 million per day of violation, and the Secretary of Energy can issue subpoenas to compel compliance. The program will run for seven years and submit reports to Congress within one year and annually thereafter with recommendations for new legislation related to AI nuclear safety. Information submitted by participating companies is kept confidential except when required by court order, Congressional request, or national security concerns.

Add a comment

Sign in to comment on this bill.

Annotate the text

Highlight any passage on the Summary or Full Text tab to attach a note. Annotations appear on the Annotations tab.