VA's AI Deployment Outpaces Safety Protocols, OIG Warns

VA clinicians using AI tools without adequate safeguards

The Department of Veterans Affairs is deploying artificial intelligence faster than its safety infrastructure can support, according to a recent Office of the Inspector General investigation. The watchdog found that VA clinicians are routinely using AI chatbots for patient care tasks without the risk management protocols the agency requires for high-impact clinical tools.

The OIG released an advisory to VA physicians in February based on preliminary findings, signaling concern serious enough to warrant immediate action. The full report, first detailed by U.S. Medicine, reveals a gap between VA's stated AI governance framework and actual practice on the ground.

Why it matters

As healthcare systems race to adopt generative AI, the VA investigation exposes a critical vulnerability: tools designed for general productivity are being repurposed for clinical decisions without corresponding safety measures. With 68% of VA's 253 health-related AI use cases classified as high-impact, the agency's inability to track AI-generated errors or identify AI-created medical records represents a systemic blind spot that could affect patient safety at scale.

Classification mismatch creates oversight gap

VA GPT and Microsoft Copilot Chat are both widely available to VA staff but have not been designated high-impact tools. Yet the OIG investigation found extensive clinical use. Among 135 AI prompts shared on an internal VA Teams channel, 79 were clinical in nature—56 for clinical notes, 17 for summarization, and six for other purposes.

By contrast, VA's Ambient AI Scribe, which drafts medication notes from clinical visits, carries a high-impact designation and includes predeployment testing, ongoing monitoring, human oversight, and feedback mechanisms. The chatbots lack these safeguards despite similar clinical applications.

Research on generative AI in medicine has documented that poor prompting technique can produce hallucinations—instances where AI systems present false information as fact. When this output enters patient records, it can compromise diagnosis and treatment decisions.

No system to track AI-related medical errors

The investigation found that VA's Joint Patient Safety Reporting system, which captures medical errors and near misses, has no AI-specific labeling process. A search by OIG investigators found zero mentions of AI in the system.

"Without a way to tag or trace AI-generated documentation, VA cannot readily detect patterns, investigate AI-generated safety events or implement quality improvement processes that may lead to safer prompting," the report stated.

VA also lacks a method to identify records created with AI, preventing retroactive review of potential safety concerns. One National Center for Patient Safety AI lead acknowledged the challenge: "It's very tricky. We don't have the best answer yet."

Policy context: aggressive deployment directive

VA leaders have testified before Congress about plans to "aggressively deploy" AI technology, aligning with April 2025 Office of Management and Budget guidance directing agencies to "remove unnecessary and bureaucratic requirements that inhibit innovation and responsible adoption."

One VA AI leader told OIG investigators this guidance has been interpreted as a directive to "push risk management down to the lowest reasonable level."

As of January 2026, VA had 367 AI use cases in its inventory, with 253 within Veterans Health Administration.

VA response

In response to the findings, VA is working with the Defense Health Agency to enable tagging of AI-related events in safety reporting systems. The agency has also increased communication between AI programs and the National Center for Patient Safety, which plans an education campaign to encourage recognition and reporting of AI-related patient safety events.

The details were first reported by U.S. Medicine based on the VA Office of the Inspector General report.