Bipartisan Senate bill calls for Energy, NIST team-up on AI evaluations

A bipartisan group of senators is making another run at a bill aimed at improving the federal government’s ability to test and evaluate artificial intelligence systems.

The Testing and Evaluation Systems for Trusted Artificial Intelligence Act — which advanced out of the Senate Commerce Committee last year but didn’t receive a floor vote — calls on the National Institute of Standards and Technology and the Department of Energy to create a testbed pilot program for the development and refinement of measurement standards used to assess AI systems.

The five lawmakers behind the legislation — Sens. Ben Ray Luján, D-N.M., Marsha Blackburn, R-Tenn., Dick Durbin, D-Ill., Jim Risch, R-Idaho, and Peter Welch, D-Vt. — see the DOE’s national laboratories playing a crucial role if the TEST AI Act is enacted. Luján, Blackburn, Durbin, Blackburn and Risch are co-leads of the Senate National Labs Caucus.  

“AI has reached every sector in our country and driven innovation, but we cannot ignore the vulnerabilities and risks that come with it,” Luján said in a statement shared with FedScoop. “While these systems have the power to change lives, they can also fall short — providing inaccurate or biased data — and are at risk of malicious attacks or misuse by our adversaries.”

“The TEST AI Act addresses these shortcomings by creating government testbeds to better evaluate AI systems,” added Luján, a frequent advocate for New Mexico’s Los Alamos and Sandia national labs. “This will help leverage the talent of our National Laboratories and strengthen the federal government’s ability to implement responsible guardrails that protect our national security and the American people.”

Blackburn, who has championed quantum legislation that builds on work done at Tennessee’s Oak Ridge National Laboratory, said the TEST AI Act would direct NIST, DOE and national labs teams “to establish safeguards, enabling AI to evolve while lowering the risk of manipulating this technology.”

The bill, shared first with FedScoop, seeks to not only codify the NIST-DOE partnership on the evaluation of AI models, but also aims to boost public-private partnerships via an AI-testing working group. That group, made up of no more than 10 members, would include the top NIST, Energy and Commerce officials or their designees, as well as representatives from the private sector and academia.

That body would guide the development of standards for AI testing, with reliability, performance, capability, interpretability, security, privacy and data bias in mind. A strategy for measurement standards would then be posted on a Commerce Department website.

The first testbed would be detailed in a report to Congress within 180 days of the initial demonstration. Additional recommendations for future standards development would also be included in the report. 

The legislation has at least one prominent non-governmental backer in its corner: the nonprofit Americans for Responsible Innovation, which has pushed the government to modernize federal systems and warned the Trump White House that cuts to NIST would put the country behind the eightball in the global AI race. 

“The TEST AI Act is a step towards transparency and accountability in artificial intelligence,” ARI President Brad Carson said in a statement. “Right now, AI systems are being deployed in high-stakes environments without independent oversight or clear standards. By building federal capacity for rigorous AI evaluations, this bill helps ensure AI tools are secure, effective, and ready for deployment.”