GAO urges agencies to share AI buying lessons
Contracts shape testing and vendor changes.
- Event / announcement date
- 2026-04-13
GAO’s April 13 report examined 13 AI acquisitions at Defense, Homeland Security, Veterans Affairs and the General Services Administration. It urged systematic collection and sharing of lessons on data rights, testing and costs. The selected case sample was not a nationwide failure rate. [1]
Earlier OMB guidance, issued April 3, 2025, explains what contract safeguards can do: data and model portability can reduce the cost of changing vendors. Contracts must allow continuing performance evaluation; independent evaluations should use agency-defined data unavailable to vendors. Buying a tool therefore also means securing the ability to scrutinize and replace it. [2]
GAO’s GSA example makes that concrete: officials needed a way to test whether newer, more expensive model versions justified their price. Its recommendations are not proof that agencies subsequently implemented them. [1]
The source
- GAO: AI acquisition lessons ↗
Source date 2026-04-13 · reviewed October 29, 2026 - OMB: AI acquisition memorandum ↗
Source date 2025-04-03 · reviewed October 29, 2026
About this item: Retrospective report. No first-hand reporting or independent testing is claimed. Corrections policy.
