PRIA Logo™
119-hr-9965119th Congress

AI Threat Output and Monitoring Incident Containment Act

Sponsored By: Representative Maloy, Celeste [R-UT-2]

Introduced

Summary

Evaluate and reduce the risk that powerful AI could cause or enable a nuclear incident. This bill creates a Department of Energy program, run through the National Laboratories, to test AI systems for nuclear-related risks and develop containment and mitigation guidance.

Show full summary
  • Large AI developers must participate if they meet the bill's threshold (about $2 billion spent on AI over the prior five years). They must give secure access to models and testing data or face subpoenas, DOJ referrals, and civil penalties up to $1 million per violation.
  • The Department of Energy and the National Laboratories will run red-team and blind evaluations, deliver a testing report for each participant, and issue recommendations on containment protocols, contingency planning, and technical standards. The Secretary must stand up the Program within 90 days and the Paperwork Reduction Act does not apply.
  • Information submitted is generally exempt from Freedom of Information Act requests but can be shared for court orders, congressional requests, national security or public safety, or as anonymized data. The Secretary must report legislative recommendations to Congress within 1 year and update them annually until the Program ends after 7 years.

Personalized for You

How does this bill affect your finances?

Personalize government policy and PRIA will tell you what this bill means for your household, plus every other piece of legislation we track. PRIA reads each provision against your financial profile to show you exactly what matters to your wallet.

Free to download

Bill Overview

Analyzed Economic Effects

5 provisions identified: 3 benefits, 1 costs, 1 mixed.

New AI nuclear testing program

If enacted, the bill would require the Energy Secretary to start an "Advanced Artificial Intelligence Nuclear Evaluation Program" within 90 days. The Program would test AI for the chance of an AI-related nuclear incident and use red-teaming and practical jailbreaking tests. It would allow outside third-party and blind evaluations when practical. Each participant would get a report with testing results, identified risks, and recommended safety steps. The Program would also develop containment, contingency, and mitigation guidance and update Department Directives for National Laboratories. The Program would end seven years after enactment.

Secretary reports and recommendations

If enacted, the bill would require the Secretary to assess Program data and report to Congress within one year. The report would identify trends, evaluation awareness, potential for AI nuclear incidents, and scheming behavior. The Secretary would recommend legislative or regulatory steps, which could include funding, licensing or certification, a new federal office or agency, or regulatory oversight. The Secretary must update the recommendations one year after the initial report and then annually until the Program ends.

Who the AI program covers

If enacted, the bill would define which developers and AI systems the Program covers. It would treat an entity as a "large advanced artificial intelligence developer" if it and its affiliates spent $2,000,000,000 or more on AI development in the prior five calendar years. "Advanced artificial intelligence" would include models trained with more than 10^26 operations unless the Secretary sets a different threshold. The bill would define a "loss-of-control" as at least 5 deaths, at least 50 serious injuries, or more than $100 million in damages. A change that costs $5,000,000 or more and materially alters AI performance would count as a substantial modification. The Program would end seven years after enactment.

New penalties for large AI developers

If enacted, the bill would require very large AI developers to participate in the Program and lists banned conduct. Prohibited actions include failing to participate, lying about participation, hiding material information, impeding the Program, refusing secure access to models (including weights and un-mitigated versions for testing), and destroying or altering records to impede the Program. After notice and an opportunity for a hearing, the Secretary could fine up to $1,000,000 per violation and treat each day of ongoing noncompliance as a separate violation. The Secretary could also issue subpoenas for model weights, software, and documents and refer cases to the Attorney General for civil action.

Confidentiality rules for Program submissions

If enacted, the bill would make most information submitted to the Program exempt from public FOIA disclosure. The Secretary could disclose submissions only for a court order, with participant consent, at a congressional committee request, for national security/public safety/critical infrastructure reasons, or as anonymized data that excludes trade secrets. The Secretary must notify participants as soon as practicable before disclosures made under a court order, a congressional request, or a national security/public safety/critical infrastructure determination.

Sponsors & CoSponsors

Sponsor

Maloy, Celeste [R-UT-2]

UT • R

Cosponsors

  • Rep. Jacobs, Sara [D-CA-51]

    CA • D

    Sponsored 7/27/2026

  • Rep. Barrett, Tom [R-MI-7]

    MI • R

    Sponsored 8/20/2026

Roll Call Votes

No roll call votes available for this bill.

View on Congress.gov
Back to Legislation