SBA behind the eightball on AI use case reporting, GAO finds
The deadline for federal agencies to post their AI use case inventories in late January came and went without the Small Business Administration hitting the mark. In March, the SBA joined the fray by making its inventory public — for the first time ever.
How the agency handled the release of its 2025 inventory was emblematic of how it has generally approached federal AI compliance, according to a new Government Accountability Office report that chided the SBA for “not consistently” following reporting requirements for use of the emerging technology.
Though the SBA did finally post its AI use case inventory two months ago, requirements for agencies to publicly report their uses have been “in development since 2020,” the report noted. The Advancing American AI Act, enacted in 2022, locked in the requirement for agencies to publicly release their inventories annually.
Nevertheless, it took until this spring for the SBA to comply with the law, according to the GAO, despite the fact that the agency has used forms of the technology since at least 2021, when it deployed a machine-learning tool aimed at preventing fraud in pandemic loan programs.
That use case has since been discontinued, per the report, but Office of Management and Budget guidance “states that agencies must report not only on their current use cases, but also on those cases being piloted or not yet fully deployed, as well as those decommissioned or discontinued within the past year.”
The SBA did share with the GAO a 2024 internal inventory listing 21 use cases. Agency officials told the watchdog that each of those uses were “either decommissioned, never deployed, or paused as of March 2025.”
The pause was intended to give the agency time to review use case compliance with new executive orders, and to reflect “any change in priorities” from the SBA’s new chief information and chief AI officers. The agency’s CIO position turned over several times in the early days of the second Trump administration before landing on former fintech and IT executive Hartley Caldwell, who also serves as SBA’s CAIO.
Last month, SBA officials told the GAO that they were still updating AI-related policies to comply with EOs, while the agency’s data governance board and AI governance council were engaged in pilot projects to “test the security, value, and performance of different AI capabilities.”
“Examples include a pilot program designed to test AI’s ability to report fraud using certain SBA program data sets and various pilot or pre-pilot use cases related to IT operations and cybersecurity initiated in June, September, and December 2025,” the report stated.
The watchdog recommended that the SBA’s CIO establish policies to meet federal AI reporting requirements, in addition to documenting the implementation of those procedures as well as other key decisions. The agency concurred with the recommendation, per the report.
“By establishing policies and procedures for reporting AI use cases—including defining roles and responsibilities and documenting their implementation—SBA would be better positioned to ensure it reports its applicable AI use inventory,” the report said. “Until SBA does so, transparency will remain limited, hindering information sharing and public and congressional oversight of agency AI use.”
AI use across the federal government jumped 70% from 2024 to 2025, according to recently released OMB data, with about 3,600 use cases reported across agencies last year.
