VA’s AI use to fix disability claims backlog sparks concerns from Congress, watchdogs

The Department of Veterans Affairs is using artificial intelligence to address its massive disability compensation claim backlog, but watchdogs and Congress expressed concern about maintaining governance and staffing Monday during a House Veterans’ Affairs subcommittee hearing.

With 80% of the Veterans Benefits Administration’s 600,000 pending claims stuck in an evidence-gathering phase, Robert Orifici, the VA’s acting deputy chief information officer, said AI can help the agency make “faster and better decisions,” with a human’s help.

“Every disability claim is decided by a trained VA employee, not by AI or automation,” he said. “These tools support human decision making; they do not replace it.”

But Nikki Budzinski, D-Ill., ranking member of the House Veterans’ Affairs Subcommittee on Technology Modernization, said staff cuts to 2,700 claims examiners since January 2025 makes keeping a human in the loop all the more difficult.

“There is no world in which that does not impact the speed at which veterans receive their disability claim or the effectiveness in VA’s ability to implement a modern cloud-based IT system for VBA, and to get ahead of this talking point, the lack of personnel cannot be solved by letting the computer do the work,” she said. “In many instances, AI and automation systems make things harder for claims processors.”

Further, Budzinski said the “so-called fixes that VA has pushed, like automation and pilots of artificial intelligence tools, often produce incorrect information, compounding issues and slowing production,” which has made examiners consider themselves more of an “unpaid software tester than a claims examiner.”

“VA routinely punishes its employees for not meeting production standards, yet fails to give them reliable tools with which to do their jobs. That is unacceptable and will only lead to a further exodus of employees,” she said.

The Government Accountability Office was also concerned with how the VA is using AI, citing the agency’s past “shortfalls” with technology and concerns about AI usage outpacing governance. Witnesses before the panel said the electronic health record modernization rollout will help the claims move, too, as it makes medical record data easier to transfer.

“VA has responded to some of these challenges and made real improvements,” Sterling Thomas, GAO’s chief scientist, said about technology problems in the VA’s disability compensation program. “However, our prior work has shown that despite these efforts, VA did not consistently achieve its goals for improvement.”

The VA Office of the Inspector General has also recently put the VBA in hot water, with an April report showing 8,000 automated Pension and Fiduciary Service decisions or letters omitted favorable findings and had incomplete evidence summaries.

While Thomas also said he is “optimistic” about some of the VA’s other AI uses, like the Payment Redirect Fraud Model to detect fraudulent direct deposit changes, he said the VA “must be thoughtful about the deployment and oversight” of AI. Thomas recommended the VA use the GAO’s AI accountability framework to guide it.

“Broadly, AI is still in early stages of development and implementation, and rapid deployment without deliberate focus on governance has already led to unintended consequences,” he said. “Before pouring data science on the problem, we need solid, reliable, ground-truth data and human in the loop to ensure data reliability and application of the technology.”

Across the aisle, lawmakers and agency officials agreed that the current claims system places too much of the burden on veterans, and something must change, while not handing everything over to robots.

“Instead of a working system, the men and women who serve face an incredible burden getting the benefits they earn. They are forced to become their own private detectives, couriers, and administrators,” said subcommittee Chair Tom Barrett, R-Mich. “There must be quality assurance policies to ensure that the use of AI and automation always have a human deciding in that process.”