HHS Deploys ChatGPT to Audit $2.1 Trillion in Medicare and Medicaid Spending

On paper, the audits were always there. Every state and major grantee that spends federal health dollars files them each year, thick binders of findings that, as one senior official put it, land "with a thud" and are promptly ignored. The U.S. Department of Health and Human Services now says it has found a reader that never tires: ChatGPT.

HHS is deploying OpenAI's large language models and other AI tools to comb through the audit trail behind roughly $2.1 trillion in annual Medicare and Medicaid spending, hunting for fraud, waste and improper payments. The effort, formally the Audit Enforcement and Risk Oversight initiative, or AERO, was unveiled in May 2026, and by mid-July officials had signaled they intend to extend the same machine-driven scrutiny from Medicaid into Medicare, the program's larger and more politically sensitive half.

What HHS confirmed

According to HHS, AERO uses "next-generation AI analytical tools" to scan at least five years of audit history across all 50 states simultaneously. The department said it sent letters to every state governor and treasurer putting them "on notice," warning that chronic failure to fix flagged deficiencies could cost them federal funding. Early findings, the agency said, show some audit problems have gone unaddressed for five or more years, and that hundreds of grantees never filed required audits at all, some delinquent by more than two years.

The use of ChatGPT specifically was reported by The Wall Street Journal, which said the tool was built in part on OpenAI's models. Gustav Chiarello, HHS assistant secretary for financial resources, is leading the effort and estimates the department loses between $100 billion and $200 billion a year to waste or fraud.

"It's classic big government: Everyone files an audit and it lands with a thud and no one does anything about it," Chiarello told the Journal. "Here, with AI, we're able to dig into it."

How it grew

The Journal reported that AERO grew out of Chiarello's review of child care fraud in Minnesota and that HHS is now scrutinizing money flowing through major universities to subgrantees. Federal law already requires any state, local government, nonprofit or university spending $1 million or more in federal funds annually to file a so-called Single Audit; AERO's premise is that the government has simply never had the capacity to read them all. Potential enforcement actions range from temporarily withholding payments and disallowing costs to suspending awards and initiating debarment.

It arrives amid a wider federal fraud push that has accelerated through 2026, from Medicaid funding deferrals to Medicare enrollment freezes and a new requirement that all 50 states audit their Medicaid providers. Where Medicaid runs through the states, Medicare operates under a separate framework; Chiarello has said oversight there will expand "as the program develops," a caveat worth remembering when the $2.1 trillion figure is invoked.

The accuracy and privacy problem

What HHS has not released is arguably more consequential than what it has. As of publication, the department has not disclosed the AI system's error rate, its methodology, its training data, or a timeline for when funds might actually be cut, and the letters to states carried no deadline for corrective action. That silence is the crux of the criticism.

Healthcare provider groups warn that overly aggressive algorithms produce false positives that can freeze legitimate payments and disrupt care for vulnerable patients. The American Hospital Association has called for transparency in how the models are trained and validated. Privacy and legal experts have separately flagged that HHS's public materials say little about how HIPAA-protected data on millions of law-abiding beneficiaries will be secured, a pointed concern given that OpenAI itself has disclosed data-exposure incidents. Some attorneys are already advising grantees that receive AERO letters to file Freedom of Information Act requests seeking the AI's validation studies and bias assessments before responding.

The deeper tension is one of governance, not technology. A chatbot flagging a suspicious pattern is a starting hypothesis; using that flag to withhold funding from a hospital or state agency is an enforcement decision that carries due-process obligations. The distance between those two acts is exactly where AERO has, so far, published the least. Automation can make an under-resourced oversight system faster without making it fairer, and at $2.1 trillion, the cost of being confidently wrong scales with it.

What to watch

Watch for HHS to publish (or refuse to publish) AERO's error rate and methodology; for the first concrete funding withholding, which will trigger the initial legal challenge and test whether an AI-derived flag survives judicial scrutiny; and for the formal move into Medicare, which would pull the program's largest spending stream, and its enrollees' data, under the same model. How HHS answers the FOIA requests now landing on its desk will tell you how much of this it is prepared to show its work on.