AI Threat Output and Monitoring Incident Containment Act

This bill would require the Secretary of Energy, working through the National Laboratories, to set up a program within 90 days to test advanced artificial intelligence for the risk of nuclear-related incidents, including attempts to get around its safety limits. Companies that have spent at least $2 billion over five years developing AI and that build or substantially modify advanced AI would be required to take part and give secure access to their AI models, including model weights. The Secretary could fine companies up to $1 million per violation, with each day of a continuing violation counted separately, and could issue subpoenas. Information companies submit would generally be kept from public release, with listed exceptions. The Secretary would report legislative recommendations to Congress, and the program would end after 7 years.

Full text and official record on Congress.gov

Sponsors

• Celeste Maloy Republican sponsor
• 2 cosponsors