In April 2026, the Government Accountability Office released report GAO-26-107859, documenting a striking pattern across four major federal agencies: AI adoption is accelerating rapidly, but the infrastructure to manage it systematically — particularly on the acquisition side — hasn't kept pace.
The headline finding: federal agencies more than doubled their documented AI use cases in a single year, growing from 571 in 2023 to 1,110 in 2024. Meanwhile, industry invested over $250 billion in AI development in 2024 alone. The scale of AI entering the federal marketplace is enormous. The processes for buying, evaluating, and overseeing it are not.
GAO reviewed AI acquisition practices at DOD, DHS, GSA, and VA — four of the largest and most active AI-purchasing agencies in the federal government. The finding was consistent across all four: agency policies did not require systematic collection of lessons learned from AI acquisitions.
This matters for a specific reason. AI products are fundamentally different from traditional software in how they perform in production versus how they perform in vendor demonstrations. An AI system that produces impressive results on a curated test dataset may behave very differently on an agency's actual data, in an agency's actual operating environment, under an agency's actual user workflows.
Without systematic lessons-learned processes, agencies are running the same risk of surprise over and over — buying AI tools, discovering the performance gap after deployment, and having no institutional mechanism to capture what happened and share it with the next acquisition team facing a similar decision.
When every AI acquisition starts from scratch — no documented evaluation criteria informed by prior deployments, no performance baselines from comparable agencies, no shared understanding of what red flags to watch for — the government is paying for the same learning curve repeatedly across hundreds of independent acquisitions.
GAO documented a second, related problem: acquisition officials across five agencies reported difficulty accessing AI technical experts — data scientists, machine learning engineers, and AI system architects — to help evaluate contractor proposals.
This creates a structural weakness in federal AI source selections. The evaluation panels reviewing AI proposals are typically staffed by subject matter experts in the agency's mission area and acquisition professionals — people who understand what the agency needs but may not be positioned to evaluate whether a specific AI architecture will deliver it reliably, securely, and at scale.
The result: source selections for AI products increasingly turn on vendor demonstrations and written proposals rather than independent technical evaluation. This shifts the advantage to vendors who are skilled at presentations, not necessarily vendors whose products will perform best post-award.
The National Defense Authorization Act for FY2026, signed December 18, 2025, layered additional requirements on top of the GAO findings. The NDAA mandated an AI sandbox task force and an AI Futures Steering Committee, both with April 1, 2026 deadlines. These structures are intended to create more systematic federal AI governance — but the implementation burden falls on agencies that are already stretched.
For procurement teams, the practical effect is more requirements, more reporting, and more oversight without a corresponding increase in contracting capacity or technical expertise. The agencies that navigate this period best will be those with acquisition infrastructure — systems, processes, and tools — that can absorb the additional compliance requirements without grinding day-to-day procurement to a halt.
The GAO report is a clear-eyed diagnosis of a real problem: federal AI adoption is accelerating faster than the acquisition infrastructure to support it. The agencies that close that gap — building systematic evaluation, documentation, and oversight processes — will get better value from AI investments and reduce post-award performance risk. The agencies that don't will keep paying for the same surprises.
ArcSuite AI provides the audit trail, evaluation structure, and systematic documentation that GAO says agencies are missing. Available on GSA Schedule.