AI Threat Output and Monitoring Incident Containment Act

Key Points

• The bill creates a new Advanced Artificial Intelligence Nuclear Evaluation Program run by the Energy Department through the National Laboratories to test whether powerful AI systems could help someone build or misuse nuclear weapons or interfere with nuclear facilities.
• Only the largest AI companies are covered: those that have spent at least $2 billion combined on computing power, data, personnel, and research over the past five years must join the program and hand over access to their systems, including unreleased versions without safety filters.
• Companies that refuse to participate, lie to investigators, or hide information face civil penalties of up to $1 million per day of violation, and the Energy Secretary can subpoena their AI models and refer serious cases to the Attorney General for a lawsuit.
• Test results will be used to write new safety rules for the National Laboratories, and within one year the Energy Secretary must tell Congress whether new laws, licensing rules, or even a new federal agency are needed to prevent an AI-related nuclear disaster.
• The program is set to run for seven years, and while most company data submitted for testing is shielded from public records requests, the Secretary can still share it with Congress, courts, or the public if there is a serious national security or safety concern.