AI Case Studies  /  Case study

How U.S. Federal Agencies Are Actually Using AI: Inside the Inventory and the New Rules

A look at what federal agencies have publicly documented they are doing with AI, from veteran cancer screening to internal chatbots, and how OMB's 2025 memos rewrote the governance and procurement rules that now sit under every one of those uses.

By Anthony Guerriero  ·  Reviewed by The Leveraged Years Editorial Desk  ·  Published June 28, 2026

In 2024, 37 federal agencies publicly reported more than 1,700 AI use cases to the Office of Management and Budget. The biggest categories were internal mission support, health and medical work, and government service delivery. The Department of Veterans Affairs, the Department of Health and Human Services, and the Department of Homeland Security account for some of the most documented uses, from computer vision that assists cancer screening to internal generative AI chatbots. Since April 2025, all of those uses operate under OMB memos M-25-21 and M-25-22, which require each agency to name a Chief AI Officer, publish an AI strategy, and apply minimum risk controls to "high-impact" systems.

What the inventory actually shows

More than 1,700. That is how many publicly reportable AI uses 37 federal agencies disclosed for 2024, and it is the single clearest number we have for how far AI has spread inside the U.S. government. The count comes from the annual use case inventory that agencies submit to the Office of Management and Budget, which then consolidated the 2024 entries and posted them to its public GitHub repository. The figure is roughly triple the 2023 inventory. The three biggest categories were internal mission support, health and medical work, and government service delivery, the benefits-and-services work that touches the public most directly.

The Government Accountability Office put a tighter frame around the same trend. In its July 2025 report, GAO-25-107653, GAO reviewed 11 selected agencies and found their reported AI use cases nearly doubled, from 571 in 2023 to 1,110 in 2024. Generative AI alone rose about nine-fold over that period, from 32 documented uses to 282. GAO left the Department of Defense out of the inventory portion of its review because DOD is exempt from certain public reporting requirements. The story holds across every source: adoption is real, it is speeding up, and generative tools are the fastest-growing slice of it.

Health and veterans: where the heaviest uses live

The Department of Veterans Affairs is one of the most documented adopters, in large part because so much of its AI sits close to clinical care. VA's 2024 inventory lists more than 200 initiatives, and the agency classified a majority of them as safety- or rights-impacting, a sign of how near to the patient those tools run. Here is one concrete example. VA's Veterans Health Administration has put FDA-approved devices into service that use computer vision to assist clinicians during colonoscopy. A VA study reported that the technology was tied to a statistically significant 21 percent increase in the odds of adenoma detection, and an absolute detection-rate increase of about 4 percent compared with colonoscopy without the device. Higher adenoma detection rates are associated with lower late-stage cancer incidence and reduced mortality.

VA has also been honest about where it is moving slowly. In its 2024 inventory, the office running VA's new electronic health record limited its documented AI use to a single "initiated" case: a tool called Machine Algorithm for Report Surveillance, or MARS, meant to triage ServiceNow incident tickets and flag potential patient-safety impact. That detail, reported by Nextgov/FCW, is a useful corrective to the idea that agencies are rushing AI into their most sensitive systems. Many are staging it deliberately.

The Department of Health and Human Services tells a parallel story. HHS reported 271 AI uses in 2024, a 66 percent jump from its 163 in 2023, according to FedScoop. The mix is tilting toward internal generative AI. HHS's ChatCDC tool, powered by Azure OpenAI large language models, is built for internal staff on tasks such as pulling information out of text, generating software code, and recommending edits to documents. Worth noting: fewer than one percent of HHS's reported uses were classified as high-impact. Against the sheer volume of the inventory, that detail shows how much of federal AI is back-office work rather than public-facing decision-making.

Security, borders, and law enforcement

At the Department of Homeland Security, AI tilts toward operational and enforcement work. DHS reported that 36 percent of its inventory, 86 use cases, supports law enforcement. Inside that group, Customs and Border Protection documented its ICAD Automated Item of Interest Detection, which uses Matroid software to process and annotate images from field imaging equipment and decide whether they contain human subjects, drawing on trained computer vision models to recognize objects, people, and events. DHS keeps its own public AI use case inventory and an inventory library, both of which give a more granular view than the consolidated OMB file. These are the uses that draw the most outside scrutiny, because computer vision in an enforcement setting carries clear rights implications, which is one reason the governance rules below matter so much.

GSA: the platform and the in-house pilots

The General Services Administration plays two roles. It is an adopter, and it is increasingly the plumbing other agencies build on. On August 14, 2025, GSA launched USAi.gov, a no-cost platform that lets federal agencies experiment with commercial AI models, including offerings reported to involve providers such as Google, Meta, and Anthropic's Claude. GSA positioned USAi as a way to advance the White House "America's AI Action Plan" and to give agencies a governed sandbox instead of ad hoc procurement.

GSA's own 2025 use case inventory shows the texture of practical, internal adoption. Documented uses include the Leasing Desk Guide Bot, a retrieval-augmented chatbot that helps staff pull accurate answers from the Leasing Desk Guide, and PBS AI Chatbot Domain Enhancements, which gather scattered Public Buildings Service policy documents into a citation-rich assistant. The National Computerized Maintenance Management System chatbot offers contextual support inside IBM Maximo for the roughly 20,000 facility work orders submitted daily, and a ServiceNow virtual agent named Curie helps employees resolve IT issues. None of these is glamorous, and that is exactly the point. Most documented federal AI is internal productivity work: drafting documents, summarizing notes, generating first-draft code, and retrieving policy guidance.

The rules underneath all of it

Every use above now runs under a governance regime that was rewritten in 2025. On April 3, 2025, OMB issued two memos: M-25-21, "Accelerating Federal Use of AI through Innovation, Governance, and Public Trust," and M-25-22, "Driving Efficient Acquisition of Artificial Intelligence in Government." Both put into practice Executive Order 14179, "Removing Barriers to American Leadership in Artificial Intelligence," signed January 23, 2025. They replaced the prior Biden-era guidance and shifted the emphasis toward faster adoption while keeping a governance backbone in place.

The operational requirements are specific. Under M-25-21, every executive agency must name a Chief AI Officer within 60 days if it had not already done so, and within 180 days must develop an AI strategy for identifying and removing barriers to responsible AI use and improving maturity across the enterprise. For what it deems "high-impact AI," M-25-21 sets minimum risk management practices, including pre-deployment testing, AI impact assessments, and ongoing monitoring. M-25-22 governs how agencies buy AI. It pushes them toward U.S.-developed solutions where possible, requires tracking of AI performance to protect taxpayer dollars, and stresses cross-functional acquisition. Its requirements apply to contracts awarded under solicitations issued on or after September 30, 2025, and to renewals or extensions exercised on or after October 1, 2025. For contractors, those two dates are the line that separates the old rules from the new ones.

This is the structure to keep in view. The inventory tells you what agencies are doing. The GAO report tells you how fast the curve is bending. The OMB memos tell you the controls and procurement terms that now sit under every documented use. Read together, they describe a federal AI program that is broad, still mostly internal, and newly bound by a named-accountability and risk-testing regime.

This is analysis of public federal records, not legal, compliance, or procurement advice.

Browse all AI Case Studies

This is analysis of public federal records, not legal, compliance, or procurement advice.