AI Threat Output and Monitoring Incident Containment Act

The bill directs the Secretary of Energy, through the National Laboratories, to establish the Advanced Artificial Intelligence Nuclear Evaluation Program within 90 days of enactment. This program is designed to test and evaluate advanced artificial intelligence to assess the likelihood of an AI nuclear incident, which encompasses scenarios like generating information for unlawful nuclear development or loss of control over nuclear systems. Key activities include implementing protocols to address jailbreaking techniques, such as red-teaming by experts simulating malicious actors, and facilitating independent, third-party assessments of AI models.

Participation in this program is mandatory for large advanced artificial intelligence developers, defined by significant investment in AI development, who must provide secure access to their AI, including model weights, for testing. Based on the evaluation data, the program will develop recommendations for containment protocols, contingency planning, and mitigation strategies for AI nuclear incidents and loss-of-control scenarios. These recommendations will also inform policies and technical standards for the National Laboratories.

The legislation includes strict prohibitions against non-compliance, such as failing to participate, making false statements, or impeding the program, with potential civil penalties of up to $1 million per violation and the possibility of subpoenas for non-cooperation. Information submitted to the program is generally confidential, with specific exceptions for disclosure related to national security or legal processes. Furthermore, the Secretary is required to submit annual reports to Congress with legislative recommendations to prevent AI nuclear incidents, and the program is set to terminate after seven years.