TEST AI Act of 2025
The Testing and Evaluation Systems for Trusted Artificial Intelligence Act of 2025 (TEST AI Act of 2025) would require the Director of the National Institute of Standards and Technology (NIST) to create a pilot program that uses testbeds to develop measurement standards for evaluating AI systems used by federal agencies. Working with the Department of Energy (DOE) and under a new Artificial Intelligence Testing Working Group, the bill aims to establish an iterative process to define how AI systems should be measured for reliability, safety, security, privacy, bias, interpretability, and applicability. It also requires a collaborative memorandum with DOE, a public strategy for measurement standards, and a formal reporting process to Congress after the first testbed demonstration. The act emphasizes excluding citizens of certain "covered foreign countries" from participation in the working group. Key elements include defining testbeds, creating a cross-agency governance structure, publishing a standards strategy publicly, and developing demonstrable testbeds within two years, all with ongoing reviews and future legislative or administrative action suggestions.
Key Points
- 1Pilot program to use testbeds for developing measurement standards to evaluate AI systems used by federal agencies, in coordination with the Secretary of Energy and with input from a new AI Testing Working Group.
- 2Memorandum of Understanding (MOU) between the Department of Commerce (NIST) and the Department of Energy within 180 days, ensuring access to DOE resources, facilities, and cross-cutting R&D programs to support pilot activities; reassessed every two years.
- 3Artificial Intelligence Testing Working Group established within 90 days of the MOU signing, with up to 10 members including representatives from Commerce, Energy, the NIST Director, and selected private sector and academic representatives; no members may be citizens of covered foreign countries (China, Russia, North Korea, Iran).
- 4Strategy for measurement standards due within one year of the Working Group’s establishment, outlining necessary standards (reliability, performance, capability, interpretability, security including data leakage risk, privacy, data bias, and bounds of applicability), a blueprint for developing standards, initial target applications, and metrics for the pilot; the strategy must be published publicly by the Secretary of Commerce (including code and model information for developers and researchers).
- 5Development of testbeds within two years, with authority to hire experts from academia, industry, or standards organizations to support the effort.
- 6Requirement to submit a Congress-facing report within 180 days after the first testbed demonstration, covering findings, recommended revisions, needed resources, and possible legislative or administrative actions to advance measurement standards for AI systems.