Four agencies still don't require documenting AI-acquisition lessons
Summary
In June 2023, GAO found the Department of Defense had never issued department-wide guidance for buying artificial intelligence -- despite AI acquisitions already underway across the department. In April 2026, GAO reviewed a narrower but still basic requirement across four agencies: whether officials document lessons learned from AI acquisitions so other federal buyers can use them. None of the four -- DOD, DHS, GSA, or VA -- required it.
In 2023, DOD didn't even have guidance to buy AI
's 2023 review found that although numerous components were already acquiring, developing, or using AI, the department had not issued department-wide guidance for how to approach those acquisitions. was planning to write such guidance but had no concrete plans or timeline to do so, and the military services individually lacked AI-specific acquisition guidance of their own -- though officials told it would help. Some pieces existed informally: 's Chief Digital and AI Officer ran an AI marketplace called Tradewind meant to speed up procurement, and separately gathered input from 13 private-sector companies on how they successfully buy AI. But none of it amounted to the department-wide framework said was needed.
made four recommendations in that report -- one to to prioritize department-wide guidance, and one each to the Army, Navy, and Air Force to follow with service-specific guidance once 's was in place. concurred with all four.
Three years later, a narrower ask still isn't met
's April 2026 report looked at a different, more basic piece of the puzzle: not whether agencies had a full AI acquisition framework, but whether they captured and shared what they learned from acquisitions they'd already made. In April 2025, issued guidance (M-25-22) directing agencies to update their AI policies -- including establishing ways to share knowledge and resources, such as proven contract language, through a repository was building -- by December 29, 2025. As of September 2025, before that deadline, found that none of , , , or 's policies included a formal requirement to systematically document lessons learned from AI acquisitions, even though some of those same policies acknowledged the value of doing so.
built its review around a nongeneralizable sample of 13 AI acquisitions across four agencies -- four each from and , three from , and two from -- analyzing 44 contracts and agreements awarded between September 2018 and February 2025. A fifth agency, Commerce (home to NIST), was interviewed about acquisition challenges and knowledge-sharing readiness but wasn't part of the 13-acquisition sample or the recommendations.
View data as table
| DHS | 4 |
|---|---|
| DOD | 4 |
| VA | 3 |
| GSA | 2 |
The gap showed up in concrete missed opportunities. retired an AI tool called SoKAT that supported suicide-prevention efforts in January 2023 after deciding it didn't justify its cost -- without documenting why, even though runs other AI acquisitions on the same topic that could have used that knowledge. didn't document that a geospatial damage-assessment AI model struggled to tell different types of dwellings apart, even as the agency extended that vendor's contract for another 12 months. drew its own parallel to a separate finding: had also not fully implemented its category-management policy in response to guidance, missing opportunities to save billions of dollars -- evidence, said, that agency-level follow-through, not just 's top-level directive, is what determines whether these initiatives work.
Officials across all five agencies talked to also described a common set of procurement headaches: limited access to in-house AI subject-matter experts, unclear data and intellectual-property rights in contracts, acquisition timelines built for hardware rather than software that evolves monthly, poorly defined requirements, thin testing regimes, and difficulty pricing a technology unlike anything agencies had bought before.
View data as table
| Industry AI investment, 2024 | 250 |
|---|---|
| Recent federal AI appropriation | 1.7 |
made four more recommendations in 2026 -- one each to , , , and -- to update policies so officials are required to document AI-acquisition lessons and submit them to 's repository. All four agencies concurred; specifically committed to updating its AI procurement process guide by July 2026.
View data as table
| 2023 report (DOD + 3 military services) | 4 |
|---|---|
| 2026 report (DOD, DHS, GSA, VA) | 4 |
The takeaway
- The 2023 and 2026 findings aren't identical, but they rhyme. In 2023, alone lacked any department-wide framework for buying AI. In 2026, a narrower ask -- just requiring officials to write down what they learned -- still wasn't met at or three more agencies.
- Real money is missing real lessons. didn't document why it retired an AI suicide-prevention tool even as it runs related acquisitions today; didn't document a model's known flaw even as it extended that vendor's contract.
- Both times, 's fix was accountability, not more paperwork. All eight recommendations across the two reports -- four to 's services in 2023, four to , , , and in 2026 -- were agreed to. Whether they stick is the same open question flagged with 's category-management policy: agency-level follow-through, not top-level guidance, is what has been missing.
The 2023 findings are from -23-105850, 'Artificial Intelligence: Needs Department-Wide Guidance to Inform Acquisitions' (June 29, 2023), read directly and in full via an archived copy. The 2026 findings are from -26-107859, 'Artificial Intelligence Acquisitions: Agencies Should Collect and Apply Lessons Learned to Improve Future Procurements' (April 13, 2026), also read directly and in full; it explicitly cites the 2023 report as prior work on AI acquisitions. The two reports examine related but distinct facets of the same underlying system -- -specific acquisition guidance in 2023, versus a four-agency lessons-learned documentation requirement in 2026 -- roughly 2.8 years apart.
Sources(2) ▾
- U.S. Government Accountability Office, Artificial Intelligence: DOD Needs Department-Wide Guidance to Inform Acquisitions (2023-06-29) — -23-105850, 's second report under a Senate NDAA-conference-report provision directing to review 's AI acquisition efforts -- the -only, guidance-gap facet, three years before doc-gao26-107859. Explicitly cited by doc-gao26-107859 (footnote 31) as prior work on AI acquisitions. Direct gao.gov fetch returned HTTP 403; fetched via an existing Wayback capture. gao.gov · original document
- U.S. Government Accountability Office, Artificial Intelligence Acquisitions: Agencies Should Collect and Apply Lessons Learned to Improve Future Procurements (2026-04-13) — -26-107859, a report to congressional requesters covering four agencies (, , , ) in depth plus Commerce/NIST -- the multi-agency, knowledge-sharing-gap facet, three years after doc-gao23-105850. Explicitly cites doc-gao23-105850 in its own footnotes as prior work on AI acquisitions. Fetched directly from files.gao.gov (HTTP 200); no Wayback capture needed since the mirror was directly reachable. gao.gov · original document
Comments
Always open. Logged-in readers can annotate paragraphs in place.
Federal agencies more than doubled their use of artificial intelligence from 2023 to 2024, and industry alone reportedly invested over $250 billion in AI in 2024 -- against roughly $1.7 billion Congress recently appropriated for AI efforts across the entire federal government. Two reports, three years apart, track how the acquisition side of that buildup has struggled to keep pace. In June 2023⧉, found had never issued department-wide guidance for how to buy AI capabilities at all. In April 2026⧉, checked a narrower, more basic ask across and three more agencies -- , , and -- and found none of them required staff to document what they'd learned from past AI acquisitions.