SBA’s AI fraud detection pilot didn’t include needed safeguards, OIG says

The Small Business Administration's Office of Inspector General found that the SBA failed to classify its Palantir-built AI fraud detection pilot for pandemic loans as a high-impact use case, exempting it from federal risk management standards required under OMB memo M-25-21. The OIG management advisory, released Monday, said the SBA did not list the fraud pilot on its public AI use case inventory and did not provide documentation supporting its determination that the tool — which flags loans as fraudulent — fell outside OMB's high-impact definition. The SBA suspended 111,620 California borrowers suspected of pandemic-era fraud and formalized its use of Palantir's software in July, after the OIG's review period ended in April.

SBA’s AI fraud detection pilot didn’t include needed safeguards, OIG says The federal agency that supports the growth of U.S. small businesses has been using Palantir software to identify fraud in disbursed pandemic assistance loans. The Small Business Administration has been using an artificial intelligence tool to detect fraud in its COVID-19 loan programs, but it did not label the then-pilot as a high-impact use case — a classification that requires federal agencies to meet minimum standards for risk management practices, according to SBA’s Office of Inspector General. In a management advisory publicly released https://legacy.sba.gov/document/report-26-15-sbas-process-implementing-federal-artificial-intelligence-requirements?utm medium=email&utm source=govdelivery on Monday, the watchdog noted that the agency has been piloting an AI program to identify fraud in disaster assistance loans, including the Paycheck Protection Program and Economic Injury Disaster Loans. OIG said SBA did not include the fraud pilot on the latest AI use case inventory listed on its public website, although it noted that the agency has been candid about its use of the tool. The review was conducted through April, although SBA’s public-facing inventory still does not include this use case as of the time of publication. The watchdog cited a February release https://legacy.sba.gov/article/2026/02/06/sba-suspends-111620-california-borrowers-suspected-committing-86-billion-pandemic-era-fraud , in which SBA detailed that it suspended 111,620 California borrowers suspected of committing fraud across the agency’s pandemic-era loan programs. That notice said, in part, that the agency “recently partnered with Palantir to expand the nationwide investigation into PPP and EIDL abuse.” FedScoop reported https://fedscoop.com/small-business-administration-palantir-contract-minnesota-fraud/ in January that SBA signed a contract with Palantir to provide an “SBA Fraud Prevention Pilot and Bootcamp.” SBA subsequently announced in July — after the period of OIG’s review — that it was formalizing its use of the company’s software for fraud prevention, which it said “us es advanced technology and artificial intelligence to surface data and leads, support criminal enforcement, and assist in the recovery of funds for American taxpayers.” OIG, however, said the agency did not appear to be meeting federal risk mitigation standards when it came to piloting the technology. The Office of Management and Budget released https://www.whitehouse.gov/wp-content/uploads/2025/02/M-25-21-Accelerating-Federal-Use-of-AI-through-Innovation-Governance-and-Public-Trust.pdf a memo in April 2025 calling for agencies to take steps to mitigate any potential pitfalls from their uses of AI tools deemed high impact. The watchdog’s report said that “OMB presumes any AI model that affects a person’s request for critical federal services or benefits, to include loans, fraud detection and related adjudication, is high impact.” Some of the minimum risk management practices that OMB says agencies should implement for high-impact use cases include completing an AI impact assessment, offering consistent remedies or appeals for those affected by AI-enabled decisions and providing additional human oversight, intervention and accountability for the tools’ use. Although the watchdog said it believes SBA’s use of the fraud detection tool falls under OMB’s definition of a high-impact use, investigators wrote that “when we spoke to SBA managers about this issue, they said this particular use case did not qualify as high impact, even though the model would be potentially flagging loans as fraudulent.” OIG’s report also said it “did not receive documentation from the agency supporting the determination” that the use case in question did not meet the definition of high impact. The watchdog added that, because SBA “has not convened an AI governance board, as required by OMB guidance, it has not demonstrated that its leadership and Chief AI Officer have established and overseen a formal process to determine which AI use cases meet the high-impact definition and to document those determinations.” Along with five other recommendations for SBA to shore up its compliance with federal AI requirements, the OIG called for the agency to develop a consistent process for identifying AI use cases as high impact and support its decisions with written documentation. OIG’s report is just the latest federal watchdog review of SBA’s AI uses that expressed concerns about the agency’s oversight of its tools. A May audit released https://www.gao.gov/assets/gao-26-107828.pdf by the Government Accountability Office similarly found that the agency “has not consistently reported an inventory of its AI uses.” That GAO report ultimately led to the House passing https://finstad.house.gov/2026/06/23/finstad-applauds-house-passage-of-his-legislation-the-sba-artificial-intelligence-utilization-act-of-2026/ a bipartisan measure in June that would require the SBA to report on its uses of AI and machine learning to Congress, detail the benefits and risks posed by the capabilities, the measures that can be taken to mitigate risks and how the agency “can retain human involvement in decisions informed” by the technologies. The Senate has not yet taken up the bill.