Sponsors (3)
Requires DOE to create a program within 90 days to test advanced AI for nuclear incident risk, compel large AI developers to participate, and authorizes subpoenas and civil penalties for noncompliance.
The bill increases near‑term government capacity to detect and mitigate AI-driven nuclear risks through rapid DOE testing, third‑party evaluation, and congressional oversight, but does so by compelling disclosure of proprietary models, imposing heavy penalties, and narrowing transparency and administrative safeguards—trading stronger safety oversight for business, legal, and civil‑liberties costs.
Researchers, federal employees, and the public gain faster, centralized risk assessment because DOE must stand up a program within 90 days to evaluate advanced AI for risks of causing a nuclear incident and produce actionable containment and mitigation recommendations.
Large AI developers receive detailed test reports that identify outcomes, risks, and recommended safety measures, helping developers improve model safety and reducing the chance of dangerous outputs affecting users.
Independent third‑party and blind model evaluations enable external validation and greater scrutiny of safety testing, increasing transparency and confidence among regulators and researchers.
Large AI developers must surrender model weights and unmitigated model versions for government testing, forcing disclosure of trade secrets and proprietary code.
Developers face civil penalties up to $1,000,000 per violation per day, creating massive financial exposure that could raise costs for consumers or chill investment in AI development.
Exemptions from FOIA and limits on public release reduce public transparency about tests and government handling of AI risks, limiting public oversight and accountability.
Based on analysis of 2 sections of legislative text.
Official title: To direct the National Nuclear Security Administration through the Secretary of Energy to establish the Advanced Artificial Intelligence Nuclear Evaluation Program, and for other purposes.
Introduced July 27, 2026 by Celeste Maloy · Last progress July 27, 2026
Requires the Department of Energy (through its National Laboratories) to set up an Advanced Artificial Intelligence Nuclear Evaluation Program within 90 days to test whether advanced AI systems could cause or contribute to a nuclear incident and to develop containment, mitigation, and contingency recommendations. The law compels participation by large advanced AI developers, authorizes subpoenas and reports, and allows civil penalties for obstructing or refusing to comply. The Program must support third‑party and blind model evaluations, create Department Directives for lab conduct regarding AI‑nuclear risks, implement realistic red‑teaming and jailbreak testing, and provide participants with reports identifying risks and safety measures. The Secretary may assess penalties up to $1,000,000 per violation per day and refer matters for criminal prosecution where appropriate.