GAO-26-107828, "Artificial Intelligence: Uses and Risks for Small Business Contracting and Innovation Research," went public on May 4, 2026. It is a question-and-answer style report, the kind GAO writes when it wants to lay out both the promise and the paperwork problem in the same document. The promise part is straightforward. The paperwork problem belongs to the SBA specifically, and it is not a small one.
Federal agencies are required to publicly report specific instances where they use artificial intelligence, including how each system was designed, developed and procured. That requirement traces back to Executive Order 13960 and has been reinforced since through the Advancing American AI Act. Development on the reporting rule has been underway since 2020. The SBA published its first AI use case inventory in March 2026.
Six Years, And Nobody Could Say Why
GAO asked the obvious follow-up: what happened between 2020 and March 2026. SBA officials told investigators they could not determine why the inventories had not been published, and pointed to a lack of documentation and turnover among the staff who had been responsible for AI reporting. That is the entire explanation on the record. Not a policy dispute, not a legal objection, not a resourcing fight anyone can point to. The paperwork obligation existed, the people who might have known why it wasn't being met are gone, and the files that would explain it were not kept.
This is not the agency's first pass at AI. SBA has previously used machine learning for pandemic loan fraud detection, work this site has covered before through the agency's Palantir-built fraud dragnet. In March 2025 SBA paused all of its AI initiatives for a policy review, and resumed limited pilot programs later that year. Somewhere across that timeline, a program that is legally required to tell the public what it is running went silent for six years and produced no paper trail explaining the silence.
Where This Points Next: Who Gets A Federal Contract
GAO's report is not only about the reporting gap. It is a forward-looking assessment of where SBA and other agencies' small business offices, the ones that decide which companies get access to federal set-aside contracts, might start using AI next. The upside case is real: AI could help those offices identify a broader pool of qualified small business contractors, summarize industry trends, evaluate supplier capability at scale, and cut administrative burden that currently slows every step of the process.
The risk case is the part that should worry anyone running a small business that depends on federal contracts. GAO flags market research built on incomplete data as a specific failure mode, since AI tools drawing from SAM.gov could miss capable firms that are absent from the database or carry outdated profiles, systematically excluding businesses that did nothing wrong except fall out of date in a federal system. On top of that sits automation bias, the tendency of a human reviewer to defer to whatever the algorithm recommends rather than checking it, plus the plainer failure modes of false positives, false negatives, and what GAO calls hallucinated rationales showing up inside proposal review.
Read that last phrase slowly. A hallucinated rationale in proposal review means an AI system can generate a plausible-sounding justification for accepting or rejecting a contractor that is not actually grounded in anything true about that contractor's file. That is not a hypothetical GAO invented for the report. It is the same failure mode this site has already documented in SBA's loan-side fraud detection, now flagged by GAO as a live risk on the contracting side before the tools are even fully deployed.
The Honest Case For What SBA Is Trying To Do
GAO's report does not conclude that SBA should avoid AI. The opposite, mostly. Small business contracting offices are understaffed relative to the volume of vendors and set-aside categories they oversee, and a tool that can summarize an industry or flag a wider set of qualified vendors is a genuine improvement over a contracting officer with a spreadsheet and a deadline. Nothing about wanting that capability is unreasonable, and nothing in the report treats the ambition itself as the problem.
GAO's own framing of the fix is worth repeating because it is more measured than anything this site would come up with independently: agencies need "process discipline, not technology enthusiasm." That means data quality checks before deployment, a human actually reviewing the output rather than rubber-stamping it, transparency about what the tool is doing, and compliance with the reporting law that already exists. None of that requires slowing down the technology. It requires the agency to keep the receipts, which is precisely the muscle SBA just demonstrated it had lost for six straight years.
What To Publish Before The Next Pilot Expands
- A complete accounting of every AI system SBA has used since 2020, not just the ones in the March 2026 inventory, with the gap years explained rather than shrugged at.
- A public description of how any AI tool used in small business contracting cross-checks SAM.gov data against other sources before excluding a vendor from consideration.
- Whether the House bill requiring annual AI transparency reports from SBA, H.R. 8881, becomes law, and what penalty attaches the next time an inventory goes missing.
An agency that cannot explain a six-year gap in disclosing its own algorithms is now the same agency small business owners are supposed to trust when it says its fraud-detection models and, soon, its contractor-screening models are running fairly. If an AI system ever tells you why your business did not make a shortlist, or why a loan file got flagged, and something about the reason does not add up, send us the story. The rest of what we have traced through this agency's fraud and lending side is right here.