VA discloses 367 AI use cases, 215 high-impact
The inventory is progress, but it also shows how much veteran-facing automation now depends on governance entries being true.
TL;DR
Federal Times reports VA’s 2025 AI inventory lists 367 artificial intelligence use cases, including 215 high-impact deployments across health care, benefits processing, records management, communications and internal operations. More than 50,000 personnel use commercial AI tools, and the inventory tracks testing, impact assessments, monitoring, training, fail-safes and appeal procedures for deployed high-impact systems. For veterans and clinicians, disclosure is the floor. The hard part is whether those controls catch bad outputs before they become care or benefits decisions.
Federal Times reports that the Department of Veterans Affairs’ 2025 AI inventory now lists 367 artificial intelligence use cases across the agency, with 215 classified as high-impact. The catalog reaches the parts of VA where errors are not abstract: clinical documentation, imaging and diagnostic workflows, disease-risk analysis, benefits processing, records management, customer service and toxic-exposure claims support.
The useful news is not that VA uses AI. Everyone knew that. The useful news is that VA is now publishing a more concrete map of where those systems sit and what governance artifacts are attached to them. According to the report, high-impact entries track pre-deployment testing, AI impact assessments and risk mitigation plans, independent reviews, continuous monitoring, operator training, fail-safe mechanisms and appeal procedures. That is the right checklist. It is also only as good as the evidence behind each line item.
The scale matters because VA is not experimenting in a corner office. The inventory includes Ambient AI Scribe for clinical documentation and TERA Memorandum Automation, which VA says reached 98.12% accuracy, assisted with more than 181,000 forms and saved an estimated 54,581 work hours. It also reports more than 50,000 personnel using commercial AI tools including Microsoft Copilot Chat, Microsoft Teams Premium, Grammarly and GitHub Copilot for tasks such as transcription, drafting, summarization, retrieval and software development.
That puts the Monday problem in familiar form: inventory first, control evidence second. GAO told Congress in 2025 that VA had increased its AI use cases and faced challenges with federal AI policy compliance, technical resources, budgets, AI acquisition, workforce development and securing sensitive data (https://www.gao.gov/products/gao-25-108739). A numbered inventory helps oversight. It does not, by itself, prove that model outputs are being caught before they affect a veteran’s appointment note, benefits file or appeal path.
Published ·Deep Fathom