S. 3336 — Reliable Artificial Intelligence Research Act of 2025

This section defines the following terms for purposes of the Act: (1) adversarial robustness, as the degree to which an artificial intelligence model resists attacks inducing incorrect, restricted, or harmful outputs while maintaining integrity, reliability, and privacy; (2) artificial intelligence, as defined in 15 U.S.C. 9401; (3) interpretability, as the degree to which humans understand how an artificial intelligence model makes decisions based on inputs and how outputs respond to input changes; (4) red-teaming, as a structured adversarial process testing artificial intelligence systems by simulating real-world actions to identify vulnerabilities; and (5) Secretary, as the Secretary of Homeland Security.

This section directs the Secretary of Homeland Security, not later than 270 days after the date of enactment, to commence at least one prize competition under the federal agency prize competition authority (15 U.S.C. 3719) to advance the science of interpretability relevant to commercially available or widely used artificial intelligence products. (As background, 15 U.S.C. 3719 authorizes federal agencies to award competitive prizes to stimulate mission-advancing innovation.) In carrying out the competition, the Secretary must consult with the Secretary of Commerce, the Director of the National Institute of Standards and Technology, the National Cyber Director, the Director of the National Science Foundation, and relevant U.S. artificial intelligence industry experts; may structure it into phases (e.g., submission of interpretability frameworks, interpretable models, or unique basic research) open to the same or distinct contestant pools; must evaluate submissions based on advancing broadly applicable interpretability principles, practical value for high-risk high-value use cases, and potential to create government or industry standards; and may contract with private, nonprofit, or governmental entities for design and administration.

This section directs the Secretary to commence, not later than 270 days after the date of enactment, at least one prize competition under the Stevenson-Wydler Technology Innovation Act (15 U.S.C. 3719)—which authorizes federal agencies to award prizes to stimulate mission-advancing innovation—to develop artificial intelligence models exhibiting adversarial robustness for at least one high-impact, high-risk application in government or industry. In doing so, the Secretary must consult with the Secretary of Commerce, Director of the National Institute of Standards and Technology, National Cyber Director, Director of the National Science Foundation, relevant U.S. artificial intelligence industry experts, and federal agency heads with relevant authority or expertise; may structure the competition into phases (e.g., submission of robustness frameworks, models, and red-teaming) open to the same or distinct contestant pools; must develop evaluation criteria considering advancements in broadly applicable artificial intelligence robustness principles and practical value in mitigating adversarial attacks in high-risk, high-value use cases; and may contract with for-profit or nonprofit entities or state, territorial, local, or Tribal agencies to design and administer the competition.

This section directs the Secretary to submit a report to the Senate Committee on Homeland Security and Governmental Affairs and the House Committee on Homeland Security not later than 180 days after the first prize competition concludes. The report must include (1) an evaluation of how the competition results inform the fields of interpretability and adversarial robustness; (2) an assessment of gaps in these fields identified by the Secretary during the competitions; and (3) suggested congressional actions to advance these fields and related research.