Artificial Intelligence Risk Evaluation Act

Full Title:
Artificial Intelligence Risk Evaluation Act of 2025

Summary#

This bill would create an Advanced Artificial Intelligence Evaluation Program inside the Department of Energy. The Secretary of Energy must set up the program within 90 days. The bill defines an "advanced artificial intelligence system" as a system trained using more than 10^26 computing operations, but lets the Secretary propose a different definition that would take effect only if Congress approves it.

The bill requires companies that design, build, or substantially modify such advanced AI systems to participate in the program and to give the Secretary materials on request. Materials may include source code, training data, model weights, deployment interfaces, and detailed training or architecture information. The bill bars anyone from deploying an advanced AI system in interstate or foreign commerce unless they follow the participation and information rules. A violation carries a fine of at least $1,000,000 per day.

The program must run standardized and classified testing and evaluations, including red-team adversarial testing and blind third-party assessments when feasible. It must provide formal reports to participants, develop containment and mitigation strategies, and inform evidence-based standards and regulatory options. The Secretary must send Congress a detailed plan within 360 days that summarizes test outcomes, recommends standards and oversight structures, proposes monitoring approaches for hardware and computation, and assesses the potential for systems to exceed human oversight or approach "artificial superintelligence." The Secretary must update that plan at least once a year. The program would end 7 years after enactment unless Congress renews it.

What it means for you#

  • If you are a covered advanced AI system developer: you must take part in the program, provide requested materials (code, data, model weights, etc.), and you may not deploy covered systems in interstate or foreign commerce unless you comply. Failing to comply can trigger fines of at least $1,000,000 per day.
  • If you are a member of Congress or staff: the bill requires annual, data-based reports and a detailed plan within 360 days to inform oversight options.
  • If you are a member of the public or a non-covered entity: the bill aims to produce empirical data and risk assessments to guide federal decisions, but the bill text does not specify direct changes to services or products available to the public.

Expenses#

  • The bill specifies a civil penalty of not less than $1,000,000 per day for violations of the participation or deployment rules.
  • The bill does not specify funding amounts, appropriations, or a budget for establishing or running the program. No publicly available information on total program costs or estimated compliance costs is provided in the bill text.

Proponents' View#

The bill text states that rapidly advancing AI brings both opportunities and significant risks to national security, public safety, economic competitiveness, civil liberties, and labor markets. Supporters in the text say the United States needs a secure testing and evaluation program to produce empirical data, lessons, and insights so Congress can make oversight and regulatory decisions based on testing rather than only theory. The program is meant to identify risks and suggest containment, mitigation, and oversight options, including measures to address the potential rise of artificial superintelligence.

Opponents' View#

No publicly available information.