Think about the last time Medicaid touched your family. A kid's annual checkup. A parent in a nursing home. A prescription covered because the out-of-pocket alternative wasn't realistic. Medicaid covers roughly 80 million Americans — one in four people. Starting this May, the federal funding that runs those programs in your state has been getting evaluated, in part, by an AI audit tool built using ChatGPT.

The tool is called AERO — Audit Enforcement and Risk Oversight. The Department of Health and Human Services launched it on May 21, 2026, sending letters to all 50 governors and state treasurers warning them they were now on notice. The enforcement menu HHS published includes withholding Medicaid payments, disallowing costs, suspending or terminating awards, and initiating debarment proceedings.

No error rate for the AI has been published. No appeals process has been defined. No enforcement deadline has been set.

What AERO actually does

Let me explain what a Single Audit is, because this all depends on it.

Any organization spending more than $1 million per year in federal funds — states, counties, hospitals, universities — is required to get an annual financial audit done. These are called Single Audits. They document whether the money was spent according to federal rules. The federal government receives thousands of these audits per year. For decades, much of that paperwork sat unreviewed.

AERO is the government's attempt to actually read all of it. Think of it like a bank's fraud-detection system, but for government grant programs. When your bank flags an unusual charge, it can freeze the card. When AERO flags an audit deficiency, HHS can freeze the funding — which in practice means state agencies managing Medicaid, child care subsidies, and university research grants might suddenly find themselves in an enforcement process.

Gustav Chiarello, HHS Assistant Secretary for Financial Resources, is leading AERO. The Wall Street Journal reported the tool was built in part using ChatGPT, alongside other large language models (AI programs trained to read and analyze text). The system scans at least five years of audit history — looking for repeat deficiencies, material weaknesses (accounting-speak for a control failure significant enough to affect financial reporting), delinquent submissions, and chronic noncompliance.

Chiarello said early findings showed some audit deficiencies going unaddressed for five years or more. Hundreds of grantees hadn't submitted required audits at all — some delinquent by more than two years. The government has real problems here. That part is not in dispute.

What's in dispute is what happens when the AI gets something wrong, and nobody has said.

Source spread

• HHS Press Room — AERO announcement — hype. Official HHS framing: emphasizes fraud prevention, chronic noncompliance, five years of unresolved deficiencies.
• Healthcare Dive — HHS launches AI-backed health fraud crackdown — builder. Concrete on AERO's scope, the Single Audit framework, and enforcement powers.
• Becker's Hospital Review — HHS launches AI-powered audit crackdown on states, grantees — skeptic. Flags the broad enforcement menu and the notice letters to all 50 governors.
• TechTimes — HHS Deploys ChatGPT on Medicaid Audits: No Error Rate, No Appeals, No Deadline — skeptic. The June 25 follow-up that names the three things HHS hasn't published.

What's real and what deserves a side-eye

What's real:

• States and grantees have a genuine chronic noncompliance problem. Some audit findings sitting unaddressed for five-plus years is a real failure of oversight. AERO is trying to fix something actually broken.
• AI tools like ChatGPT are genuinely capable of reading and flagging patterns across thousands of documents. This is one of the things these systems do reasonably well. The application isn't absurd.
• The enforcement powers give HHS real leverage to get states to fix problems they have been ignoring.

What deserves a side-eye:

• There is no published error rate. An AI making consequential recommendations about federal healthcare funding — affecting programs that serve one in four Americans — should be required to say how often it gets things wrong. HHS has not said.
• There is no published appeals process. If AERO flags your state and HHS sends an enforcement notice, the path to contesting that finding is undocumented. Traditional federal enforcement comes with administrative due process built in. This program launched without specifying it.
• No enforcement deadline means no urgency to add the missing transparency. HHS can operate in "on notice" mode indefinitely while the accountability gap stays open.

What to do about it

Most people can't audit a federal audit tool. Here's what you actually can do.

• If you work in healthcare administration or manage federal grant programs: if your organization spends more than $1 million per year in federal funds, AERO is scanning your Single Audit history right now. Any audit findings that have sat unresolved for more than a year should be addressed proactively — don't wait for HHS to contact you.
• If you're on Medicaid, or your family is: you're not directly affected yet. Enforcement actions would hit the state agency first, not individual recipients. But it's worth knowing that AI is now part of the compliance picture for the program your family depends on.
• If you follow state government or health policy: check whether your state has responded publicly to the HHS notice letter. Several state attorneys general have started asking questions about due process. Knowing where your state stands is useful context.
• If this bothers you, say so to your representatives. The missing error rate and appeals process are policy gaps, not just bureaucratic oversights. These are exactly the kinds of things congressional oversight committees ask about when someone raises them. It's not pointless to write.

Further reading

• HHS Press Room — AERO: Audit Enforcement and Risk Oversight Initiative — official HHS announcement
• Healthcare Dive — HHS launches AI-backed health fraud crackdown — scope and enforcement detail
• Becker's Hospital Review — HHS launches AI-powered audit crackdown on states, grantees — notice letters and enforcement menu
• TechTimes — HHS Deploys ChatGPT on Medicaid Audits: No Error Rate, No Appeals, No Deadline — June 25 follow-up on missing transparency
• OnHealthcare — HHS Goes All-In on ChatGPT for State Audits — detailed analysis of AERO's implications for healthcare compliance