NIH expands AI pilots amid staffing reductions

The National Institutes of Health continued to ramp up its use of artificial intelligence last year, as NIH navigated a turbulent year of reorganizations and staffing cuts.

NIH’s headcount fell to roughly 17,000 employees in early 2026, down by more than 4,000 staff from just over a year ago. The agency was subject to reductions in force and probationary firings that have roiled the Department of Health and Human Services under the Trump administration.

At the same time, the NIH’s number of AI use cases increased to 124 in fiscal 2025, up from the 82 use cases tallied in 2024, according to HHS AI use case inventory data released late last month. HHS as a whole saw a 65% increase in the number of AI use cases in 2025.

NIH has taken advantage of the General Services Administration’s “OneGov” deals to access new AI products and drive more AI use over the past year, according to Nick Weber, acting director in the Office of Scientific Computing Services at the NIH Center for Information Technology.

“But I think we’ve also had to weather the storm of changes in personnel, RIFs in different parts of the organization, and everyone’s grappling with, how can we be more productive, how can we make sure we still provide the services that are needed with a diminished staff?” Weber said during a Feb. 19 event hosted by ATARC and GovExec.

“And AI has really been a boon toward being able to do that, both from the administrative and health coordination side that we do at NIH, as well as delivery in the clinical center and across the board,” Weber added. “So I think it’s both out of necessity and out of just interest and opportunity.”

The AI use cases run the gamut of NIH work, from administrative tasks like analyzing the agency’s grants portfolio to research support, lab work and clinical assistance.

Weber pointed to multiple examples, including efforts to build “domain-specific small language models” trained on NIH’s vast data sets.

“How can we build models associated with our core data sets for Alzheimer’s disease, or pick your favorite area, and how can we make those available to the broad research community, where it’s trained or fine-tuned on those data sets?” he said. “And then you can ask questions and work with things more specific to a disease domain.”

Many of NIH’s AI use cases are being piloted or are in a pre-deployment phase. Weber said the agency is working on scaling useful AI tools and services where appropriate.

“Scaling is a challenge, and there’s so many risks around that, but we’re actively looking across that spectrum that I just described, as to where and how we can use AI to enhance that experience, enhance health outcomes and achieve the mission of NIH,” he said.

NIH also has a generative AI community of practice that now includes roughly 2,000 people, according to Weber. The group meets twice a month and offers training, including for NIH leaders.

“The major thing that that did was bring it to a conversation across the organization that AI is OK, we don’t have to be scared of it,” Weber said. “Here’s people showing what they’re doing. Here’s people showing how they’re protecting data, or making sure we’re not being overreliant on the responses, have human in the loop, all of those things.”

But with 27 separate institutes and centers, Weber said coordinating AI use across NIH is a key challenge.

“So we do things like make sure for our central funding programs that we have consistent business case reviews, evaluations, and look where things can be complementary,” he said. “Maybe you could add on to this product or platform over here, rather than building a new one. That’s something that we’re trying to do to make sure we can scale.”

NIH’s Center for Information Technology is also developing communications strategies and other “playbooks” for enterprise AI rollouts.

“We have to figure out how to integrate it into all of those components, change management, other things like that,” Weber said. “It’s hard, because people want to sprint, and we have to slow them down a little bit to make sure we have good, enterprise-style support and capabilities. But I think slowing down a little is helpful to make sure we’re more successful.”

Copyright © 2026 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.