GSA, NIST team up to evaluate AI before agency deployments

The National Institute of Standards and Technology and the General Services Administration are joining forces to evaluate AI tools before agencies use them in operations.

According to a Wednesday announcement from GSA, the new partnership with NIST’s Center for AI Standards and Innovation will support its USAi platform, which allows agencies to experiment with various models. The new testing and measurement effort aims to speed up AI adoption while improving security and confidence in those systems.

“By combining GSA’s government-wide reach with NIST’s AI evaluation expertise, we’re strengthening how the federal government deploys AI,” GSA Administrator Edward Forst said in a statement included in the release. 

The announcement comes as agencies are increasingly leaning on the technology to assist workers and automate away administrative toil. But as the technology continues its rapid evolution, so too do concerns about its reliability and security. At the same time, the Trump administration has also sought to tighten its grip on AI outputs that don’t align with its policies and the government’s permissions for use of that technology. 

Per the release, the CAISI — formerly the AI Safety Institute — will provide tools and guidance to help GSA “evaluate advanced AI models, select and interpret benchmarks, and conduct hands-on testing within real federal workflows.” 

The agencies also plan to create “guidelines and checklists” that other agencies can adopt.

Notably, President Donald Trump’s AI Action Plan last year recommended building an “evaluations ecosystem” and establishing resources for agencies to conduct their own testing. 

“We’re at a pivotal time in the AI revolution and this partnership between CAISI and GSA will enable federal agencies to adopt AI in ways that help the American people,” Craig Burkhardt, NIST’s acting director, said in a written statement in the release.

While not mentioned in the release, the GSA is simultaneously floating contract language for the government’s AI deployments via its Multiple Awards Schedule.

The agency recently published a draft clause for AI system contracts that would establish ground rules for things like intellectual property rights, use of government data, security incident reporting, and rights for government evaluations. 

Per the draft, probes would “assess bias, truthfulness, safety, unsolicited ideological content, and other factors determined by the Government in order to facilitate evaluations.”

Notably, the draft guidance would require contractors to agree that their technology could be integrated with government systems for “any lawful Government purpose,” echoing the apparent impetus of a clash between Anthropic and the Pentagon that led to a governmentwide ban on the company. 

It is not clear from the partnership announcement whether the evaluations search for the type of “ideological biases” and “social agendas” targeted in Trump’s executive order aimed at preventing “woke” AI in government. 

A GSA spokesman didn’t respond to FedScoop’s requests for comment on what the current evaluation practices are, what the rubric would look like, and whether there are plans to make the assessments public.