Oversight: Advancing VA Care Through Artificial Intelligence

VA Budget and ModernizationHouse Veterans' Affairs Subcommittee on Technology Modernization · 2025-09-15 · 119th Congress
The House Veterans' Affairs Subcommittee on Technology Modernization held this oversight hearing to examine how the Department of Veterans Affairs is deploying artificial intelligence across clinical care, benefits processing, and IT operations, and to weigh the opportunities against risks to data privacy, security, and workforce capacity. Begins at 0:33:34
Transcript
Highlights

Title

VA's use of artificial intelligence in veteran health care and IT modernization

Purpose

The House Veterans' Affairs Subcommittee on Technology Modernization held this oversight hearing to examine how the Department of Veterans Affairs is deploying artificial intelligence across clinical care, benefits processing, and IT operations, and to weigh the opportunities against risks to data privacy, security, and workforce capacity. Witnesses from VA, GAO, an AI industry association, and academia testified about VA's growing AI use-case inventory, tools like the Reach Vet suicide-risk algorithm and ambient clinical documentation, and long-standing IT management challenges including the lack of a permanent Chief Information Officer. Begins at0:33:34

Who spoke

Chairman Tom Barrett (R-MI)0:33:34: Opened the hearing describing frustrations with VA technology0:33:34, later questioned Worthington on vendor data-use restrictions1:06:41 and on secure deletion of ambient-listening recordings1:54:53, and closed by stressing AI must support, not replace, clinicians1:56:09.

Ranking Member Bazinski (D-IL)0:39:44: Called AI review premature given VA's "crumbling IT infrastructure"0:39:44, criticized a proposed 20% OIT workforce cut amid calls to "hire great tech talent"0:42:25, pressed Worthington on named employees linked to DOGE and their data access1:14:06, cited a GAO/OIG finding of $7 billion in improper PACT Act payments from a flawed automated tool1:37:08, and asked for a response to Ranking Member Takano's June letter on the same data-access issue1:55:26.

Mr. Charles Worthington, VA Chief Technology & AI Officer0:46:27: Said VA's AI inventory held 227 use cases in 2024, up nearly 100 from the prior year0:48:20, that 82% of AI use cases come from VHA0:48:49, and that a generative AI tool now available to all VA employees saves users over two hours weekly0:48:20; later reported 73% of users gained more time for professional work and 68% reported higher job satisfaction1:54:07.

Mr. Sid Ghatak, National Artificial Intelligence Association0:51:17: Argued AI is broader than large language models and compared the technology's maturation to the early, accident-prone railroads0:53:48; emphasized that most AI investment chases correlation rather than causation, illustrated by a spurious correlation between Google searches for "Nintendo" and the number of Michigan librarians1:45:31.

Dr. Mohammad Ghassemi, Michigan State University0:56:54: Framed AI's value around automation, augmentation, and insight for health care delivery0:58:17, and stressed that transparency about data provenance is essential to veteran trust, comparing it to a nutrition label1:31:34.

Ms. Carol Harris, GAO1:01:59: Reported VA's AI use cases grew from 40 in 2023 to 229 in 2024, with 64% classified high-impact overall and 72% within VHA1:02:591:04:20; noted GAO has 26 open recommendations on VA's IT management1:05:57, and later flagged VA's lack of a permanent CIO and OIT's roughly $300 million funding cut and 931-staff reduction as key risks1:50:27.

Dr. Evan Carey, VA National AI Institute (Acting Director)1:10:15: Explained AI decision-support tools give providers general clinical guidance1:10:15, detailed updates to the Reach Vet suicide-risk model since 20171:34:16, and affirmed VA clinicians, not AI, remain in control of care decisions1:35:07.

Rep. Latrell1:17:15: Challenged whether VA's 229 reported AI use cases reflect all 172 VA facilities or only a subset1:18:08, questioned data consolidation efforts1:39:24, and pressed Worthington and Ghatak on who should be trusted to vet AI vendors given VA's uniquely large dataset1:21:04.

Rep. Sherless1:22:32: Asked Worthington whether VA is confident in achieving standardization for AI given past failures with electronic health records1:22:57, and pressed on how bias is addressed during model development and deployment1:24:44.

A committee member (addressed by the chair as "Mr. El")1:38:54: Questioned whether veteran data siloed across facilities could be unified into a single secure enclave1:39:24, and asked Ghatak which organizations could realistically execute such consolidation1:41:32.

Key moments

Worthington: VA's AI inventory grew to 227 use cases in 2024, up nearly 100 from the prior year, with a further update expected in December 20250:48:20.

Worthington: The Reach Vet program has used AI since 2017 to identify over 130,000 veterans at elevated suicide/overdose risk, and AI-assisted colonoscopy devices raised adenoma detection rates by 21%0:49:18.

Harris/GAO: VA's AI use cases rose from 40 in 2023 to 229 in 2024, with 64% deemed high-impact overall and 72% within the Veterans Health Administration1:02:591:04:20.

Bazinski pressed Worthington on six named employees affiliated with the Department of Government Efficiency who had access to VA data enclaves; Worthington said he could not confirm what data they accessed or whether any veteran data was removed1:14:06.

Bazinski cited an April VA Inspector General finding that VBA's automated decision-support tool caused at least $7 billion in improper PACT Act payments due to incorrect effective dates1:37:08.

Harris: OIT requested nearly $300 million less funding for FY26 and has cut 931 staff, while VA still lacks a permanent CIO — a gap Harris called "hugely problematic" for prioritizing modernization efforts including AI1:50:27.

Ghatak argued most AI models chase statistical correlation rather than causation, using the (non-causal) correlation between Google searches for "Nintendo" and the number of librarians in Michigan as an example of the risk1:45:31.

Rep. Latrell pressed Worthington on whether VA's reported 229 AI use cases were drawn from all 172 VA facilities or only a limited sample; Worthington affirmed a "comprehensive review process" without fully resolving Latrell's doubt1:18:08.

Worthington: 73% of users of VA's generative AI tool reported more time for professional skills and 68% reported increased job satisfaction, though Bazinski questioned the added burden on providers as ambassadors/monitors of new tools1:54:07.

Carey confirmed VA has no current plans to use AI or chatbots as a substitute for human providers in mental health crisis intervention, after Bazinski sought that explicit commitment1:36:03.

Metadata

CommitteeHouse Veterans' Affairs Subcommittee on Technology Modernization
Chamber / CongressHouse · 119th Congress
Date2025-09-15
TypeHearing
Witnesses
Dr. Mohammad Ghassemi — Assistant Professor, Michigan State University
Mr. Sid Ghatak — Chief Technical Advisor, National Artificial Intelligence Association
Dr. Evan Carey — Acting Director, National Artificial Intelligence Institute, Veteran Health Administration, U.S. Department of Veterans Affairs
Mr. Charles Worthington — Chief Technology Officer & Chief Artificial Intelligence Officer, Office of Information & Technology, U.S. Department of Veterans Affairs
Ms. Carol Harris — Director, Information Technology and Cybersecurity Issues, U.S. Government Accountability Office
Videoyoutube
Transcript250 caption blocks · 13,578 words · 1:58:17 runtime
EventCongress.gov 118596