TEST AI Act of 2025
- Sponsor
- Sen. Luján, Ben Ray [D-NM]
- Committees
- Commerce, Science, and Transportation Committee (primary)
- Last action
- May 7, 2025
Bottom line
The bill establishes a federal pilot program and working group to develop crucial measurement standards for trustworthy AI, leveraging interagency collaboration and expert input while addressing national security concerns.
What it actually does
This bill mandates the Director of the National Institute of Standards and Technology (NIST) to establish a pilot program utilizing testbeds to develop measurement standards for evaluating artificial intelligence (AI) systems, particularly those used by federal agencies. It requires coordination with the Department of Energy (DOE) and the establishment of an AI Testing Working Group, which will include private sector and academic representatives, to advise on strategy and identify necessary standards. The bill also prohibits citizens of certain foreign countries from serving on this working group.
Proponents argue
Supporters argue that establishing robust, transparent, and replicable measurement standards is essential for ensuring the reliability, performance, security, and trustworthiness of AI systems, especially as they become more integrated into federal operations. This initiative is seen as critical for fostering public confidence in AI, promoting responsible innovation, and maintaining U.S. leadership in the global AI landscape by providing a foundational framework for evaluation.
Opponents contend
No significant organized opposition is typically raised against the concept of developing standards for AI safety and performance. However, potential critics might express concerns about the pace of federal standard-setting keeping up with rapid AI advancements, the potential for bureaucratic inefficiencies in interagency coordination, or the cost implications of establishing and maintaining extensive testbed infrastructure without explicit appropriations.
The bill is very short and straightforward, making it highly accessible for a reader to understand its core provisions and implications within a short timeframe.
Section 3(a)
Establishment of AI Testbed Pilot Program
This section requires the National Institute of Standards and Technology (NIST) Director, in coordination with the Department of Energy (DOE) and an advisory Working Group, to create a pilot program. This program will use specialized testing facilities, known as testbeds, to assess the feasibility and advisability of developing measurement standards for evaluating artificial intelligence systems used by federal agencies. The program is designed to be an iterative process, advancing measurement science and developing standards through testing and stakeholder review.
Supporters argue
Supporters argue that this pilot program is crucial for building a robust framework to evaluate AI systems, ensuring they are reliable, secure, and trustworthy for federal use. It provides a practical mechanism to develop much-needed measurement standards in a rapidly evolving field.
Critics contend
No specific opposition is typically raised against the concept of developing standards for AI safety and performance, though concerns might arise regarding the scope or cost of such initiatives.
Tradeoffs
This provision balances the urgency of AI deployment with the critical need for rigorous evaluation and standardization, potentially introducing a necessary but time-consuming process before widespread federal AI adoption.
Section 3(b)
Interagency Coordination via Memorandum of Understanding
Within 180 days of the bill's enactment, the Secretary of Commerce (who oversees NIST) and the Secretary of Energy must sign a memorandum of understanding (MOU). This MOU is intended to ensure that NIST and other federal agencies have necessary access to the Department of Energy's resources, personnel, and facilities to support the AI testbed pilot program. The MOU must be reassessed and potentially renegotiated every two years to ensure it continues to meet the agencies' needs.
Section 3(c)
Establishment of AI Testing Working Group and Strategy Development
This section mandates the creation of an "Artificial Intelligence Testing Working Group" within 90 days after the Commerce-Energy MOU is signed. This group, limited to 10 members from government, industry, and academia, will advise on developing AI measurement standards. A key restriction is that no citizen of a "covered foreign country" (China, Russia, North Korea, Iran) may serve. Within one year, the Working Group must develop a strategy identifying necessary measurement standards (e.g., for reliability, performance, security, privacy, data bias), a blueprint for their development, initial applications, and metrics for the pilot program. This strategy must be publicly published.
Section 3(d)
Testbed Development for Measurement Standards
Within two years of the bill's enactment, the NIST Director is required to develop testbeds. This development will be done in collaboration with the Secretary of Energy and in coordination with the AI Testing Working Group, following the strategy developed by that group. These testbeds will be used to demonstrate the development of measurement standards specifically for evaluating artificial intelligence systems intended for use by federal agencies. The Director is also authorized to hire necessary experts from academia, industry, and standards development organizations to assist in this effort.
Section 3(e)
Report to Congress on Pilot Program Findings
After the NIST Director completes the first testbed demonstration and initial development of measurement standards, a report must be submitted to Congress within 180 days. This report will review the findings from the initial demonstration, recommend revisions to the pilot program plan (including identifying needed resources, funding, or personnel), and propose legislative or administrative actions to further advance the development of AI measurement standards.
The definition of 'artificial intelligence system' is not self-contained within this bill but refers to the definition provided in section 5002 of the National Artificial Intelligence Act of 2020 (15 U.S.C. 9401).
Section 2(1)
Why it matters:This is standard legislative practice to avoid redundancy and ensure consistency with existing law, rather than an attempt to obscure.
Case for: Referencing existing definitions ensures legal consistency across federal AI policy and streamlines legislative drafting, preventing conflicting interpretations of what constitutes an 'artificial intelligence system'.
Case against: For a casual reader, requiring consultation of external statutes adds a layer of complexity and makes the bill less self-contained, potentially hindering immediate comprehension of its full scope.
Estimated impact: This definition broadly encompasses any artificial intelligence as defined by the 2020 Act, which includes machine-based systems that can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments.
The bill explicitly defines 'covered foreign country' to include the People's Republic of China, the Russian Federation, the Democratic People's Republic of Korea, and the Islamic Republic of Iran.
Section 2(2)
Why it matters:This is standard legislative practice for defining terms, but its implications for the Working Group's composition are significant and could be overlooked.
Case for: Clearly defining these countries upfront ensures precision and consistency when applying the restriction on Working Group membership, aligning with broader national security objectives to protect sensitive U.S. technology and research.
Case against: While necessary for clarity, placing such a politically sensitive list in the definitions section might make its full impact on international collaboration and talent acquisition less immediately apparent to a reader focused on the main program objectives.
Estimated impact: This definition directly impacts the composition of the Artificial Intelligence Testing Working Group, excluding individuals from these four nations from participation.
The pilot program is to be carried out by building and demonstrating an 'iterative process' for advancing measurement science and developing standards using testbeds, with results reviewed by stakeholders.
Section 3(a)(2)
Why it matters:This is a detail about the program's operational approach, which is typically placed within the implementation specifics rather than in a headline section.
Case for: An iterative process is critical for developing effective standards in a rapidly evolving field like AI, allowing for continuous learning, adaptation, and refinement based on testing results and stakeholder feedback.
Case against: While beneficial, an iterative process can be perceived as slower or less decisive than a one-time standard-setting effort, potentially delaying the availability of initial standards.
Estimated impact: This provision shapes the fundamental methodology of the entire pilot program, influencing how quickly and effectively standards are developed and updated.
The strategy for measurement standards developed by the Working Group must identify standards for specific AI attributes, including reliability, performance, capability, interpretability, security (including data leakage propensity), privacy, data bias, and bounds for applicability and use cases.
Section 3(c)(3)(B)(i)
Why it matters:This is a detailed enumeration of specific technical requirements, which is often placed in subsections rather than high-level summaries.
Case for: Explicitly listing these attributes ensures a comprehensive approach to AI trustworthiness, addressing key concerns from technical performance to ethical considerations like bias and privacy, which are vital for public and federal confidence.
Case against: While comprehensive, such a detailed list could be seen as potentially prescriptive, possibly limiting the Working Group's flexibility to identify other emerging critical attributes or to prioritize based on evolving threats and needs.
Estimated impact: This provision directly dictates the scope and focus of the AI measurement standards that will be developed, influencing the entire direction of the pilot program.