HHS is launching an AI system to detect fraud and waste in government health programs
The Department of Health and Human Services is moving from “pay and chase” to real-time AI testing across Medicare, Medicaid, CHIP and the Marketplace.
The US Department of Health and Human Services has launched an artificial intelligence program aimed at detecting fraud and waste in all government health systems, building on a strategy first unveiled in February that promises to replace the agency’s “pay and chase” model with real-time reviews of claims before they are paid. Reuters reported on the development on Wednesday.
The program includes Medicare, Medicaid, the Children’s Health Insurance Program and the Health Insurance Marketplace, according to a joint HHS announcement earlier this year.
In that February release, HHS Secretary Robert F. Kennedy Jr, Vice President JD Vance and CMS Administrator Mehmet Oz pitched the change as a departure from the decades-old practice of paying claims first and investigating later to what the agency calls a “find and serve” model, using AI tools to flag suspicious claims during adjudication.
The numbers after pressing are big enough to explain the urgency. Medicare’s fee-for-service program alone generated an estimated $28.83bn in inappropriate payments by 2025, according to a CMS fact sheet; Medicare Part C added another $23.67bn.
A separate report by the Government Accountability Office in April put improper government payments worldwide at about $186bn a year, with the total concentrated in five programs, including Medicare and Medicaid.
The regulatory vehicle behind the program is a formal Request for Information that HHS and CMS opened in late February, asking for industry input on analytics methods, AI tools and data sharing methods.
The RFI closed on March 30th and is part of a proposed rule that CMS has been calling CRUSH, the “Comprehensive Rules for Suspicious Healthcare Disclosure”.
May’s move appears to be the next operational step in that discussion, although neither HHS nor CMS has published a complete list of vendors or the technical architecture behind it.
The pilots were racing in tandem. The HHS Office of Inspector General tested a machine learning model that scores providers for statistical payment behavior associated with fraud and abuse, and CMS reported that total savings to the Medicare program increased by 59% in fiscal 2025, from $26.3bn last year to $41.9bn.
The agency says part of the jump is due to enhanced vetting of new registrants, including a nationwide six-month moratorium on home health and hospice registration that went into effect on May 13.
The biggest risk of moving from post-payment review to pre-payment AI testing is what false positives do to providers. A flagged claim that delays payment in a legal process, especially a small one, is a financial deficiency event. Industry groups have already pressed CMS, through the RFI process, for clearer appeal rights and limits on human review before any objections flagged for AI become final. None of those guardrails have been written into law.
What HHS did not disclose: which vendor models are being used, whether the system will work on claims data that can be identified or not fully identified, and how the agency will evaluate the models’ error rates.
CRUSH rulemaking is the document that those answers will ultimately have to live on. At the moment, the initiative is working against the backdrop of unusually large numbers of unfair payments and the state’s desire for AI compliance that is, by recent standards, unusually high.
