Representative Maloy has introduced H.R. 9965, a bill designed to establish the Advanced Artificial Intelligence Nuclear Evaluation Program. The legislation directs the National Nuclear Security Administration, through the Secretary of Energy, to implement testing and evaluation protocols to assess the likelihood of artificial intelligence being involved in a nuclear incident.
Under the proposed act, titled the AI Threat Output and Monitoring Incident Containment Act, the Secretary of Energy would be required to establish the program through the National Laboratories within 90 days of enactment. The program's objectives include testing artificial intelligence to assess nuclear incident risks and implementing protocols to address practical jailbreaking techniques. This includes red-teaming by entities with expertise comparable to sophisticated malicious actors.
Large advanced artificial intelligence developers would be required to participate in the program. The bill defines these developers as entities that develop, own, or substantially modify advanced artificial intelligence for interstate or foreign commerce and have collectively expended at least $2,000,000,000 on AI-related investments over the previous five calendar years.
To ensure compliance, the Secretary may issue subpoenas to compel the production of artificial intelligence, including model weights and related documentation. Violators could face civil penalties of up to $1 million, with each day of a continuing violation constituting a separate offense. The Secretary also maintains the authority to refer matters to the Attorney General for civil action in a U.S. district court.
The legislation includes specific protections and definitions regarding AI nuclear incidents. An AI nuclear incident is defined to include scenarios involving the generation of technical information likely to assist in the unlawful development or acquisition of nuclear weapons or material, or the generation of Restricted Data. It also covers loss-of-control scenarios where an AI pursues objectives different from those intended by a human developer, resulting in specific thresholds of death, injury, or economic damage.
The program would also focus on identifying trends and risks, such as scheming behavior and evaluation awareness. Based on these findings, the Secretary is expected to provide recommendations for future legislation, which could include new regulatory oversight, licensing processes, or the establishment of a new federal agency or office.
The proposed program is slated to terminate seven years after the date of the act's enactment. The bill has been referred to the Committee on Science, Space, and Technology.
