(Just The News)—As the federal government faces hundreds of billions of dollars in estimated annual fraud losses, a bipartisan majority of voters want it to use artificial intelligence to help catch waste and abuse, a new poll found.
In the latest The Center Square Voters’ Voice Poll, 60% of registered voters said they support using artificial intelligence to help detect fraud, waste or abuse in government programs such as Medicaid. Twenty-seven percent opposed it and 13% were unsure.
The federal government could lose an estimated $233 billion to $521 billion a year to fraud, according to the Government Accountability Office, based on fiscal 2018-2022 data, though the tools meant to catch it have a mixed record.
Support crossed party lines. Nearly three-quarters of Republicans, 74%, backed the use of AI to detect fraud, along with 57% of independents and 48% of Democrats. Even among Democrats, the group most divided on the question, more supported the idea than opposed it.
The question comes as the government expands its use of AI and data matching to flag suspect payments. In fiscal 2025, federal agencies reported an estimated $186 billion in improper payments, about 82% of it from overpayments, concentrated in five programs that include Medicare and Medicaid, the program the poll asked voters about.
The Centers for Medicare and Medicaid Services screened 1 billion payments through the Treasury Department’s Do Not Pay system in fiscal 2024 and prevented 363,386 payments worth $2.6 billion.
Of $765.8 billion in overpayments that agencies attributed to data-access problems between fiscal 2021 and 2024, about 73% — or $556.6 billion — stemmed from a failure to access data agencies already possessed, according to the Congressional Research Service.
The scale of outright fraud has been more difficult to measure. The GAO estimate, from 2024, was the first and only government-wide fraud figure of its kind. GAO has said it has no plans to update the spending-side number, having recommended that the Treasury Department develop a method for producing one instead. Treasury had not done so as of July 2026.
Romina Boccia, director of budget and entitlement policy at the Cato Institute, said AI can help but is no substitute for controls agencies already have. AI “cannot compensate for weak eligibility verification, incomplete audits, poor data sharing, or a lack of accountability when agencies and contractors make repeated errors,” she told The Center Square.
Boccia said the technology should be judged by whether it “helps administrators prevent improper payments before they go out, produces verifiable savings, and does so without significantly increasing the denial rate of eligible people to receive benefits or necessary care.”
The AI tools are not foolproof. Of the 48 agencies that used the Treasury Department’s Do Not Pay system in fiscal year 2024, 15 reported it was not effective at preventing improper payments for them, according to the Congressional Research Service. The Defense Department reported that “the majority” of the matches the system flagged were false positives, such as a vendor’s tax identification number matching a deceased person’s Social Security number.
In 2024, a federal judge ruled that Tennessee’s automated Medicaid eligibility system had unlawfully terminated coverage for thousands of residents, finding it violated their due-process rights by relying on faulty data and incorrect household determinations.
Federal law limits how far a data match can go on its own. Under the Computer Matching and Privacy Protection Act, an agency generally cannot cut off someone’s benefits based solely on a data match. It must independently verify the match and give the person notice and an opportunity to respond. Those protections remain in place.
Using authority Congress granted in 2019, the Treasury Department in 2025 issued a four-year waiver allowing qualifying Do Not Pay matching programs to bypass the formal data-matching agreements and cost-benefit analyses agencies would otherwise have to complete. That came after a Trump executive order directed it to reduce administrative barriers.
The push to streamline that review has drawn objections. The Electronic Privacy Information Center, a privacy research group, argued in an Aug. 11 comment to the Department of Health and Human Services that comparing TANF recipients’ personal data against other databases without the required published matching agreements would violate the Computer Matching and Privacy Protection Act. The administration has said the change is meant to speed the detection of improper payments.
The Center Square Voters’ Voice Poll was conducted by Noble Predictive Insights from Aug. 12-16, surveying 2,533 registered U.S. voters. The sample included 930 Republicans, 930 Democrats, and 673 Independents. Among independent voters, 330 respondents were classified as True Independents, or those who do not lean toward either major party when given the choice. The margin of error is plus or minus 2.0 percentage points.
