The Department of Health and Human Services is moving from “pay and chase” to real-time AI screening across Medicare, Medicaid, CHIP and the Marketplace.
The US Department of Health and Human Services has launched an artificial intelligence initiative aimed at detecting fraud and waste across federal health programmes, building on a strategy first outlined in February that promises to replace the federal “pay and chase” model with real-time screening of claims before they are paid. Reuters reported the development on Wednesday.
The programme covers Medicare, Medicaid, the Children’s Health Insurance Programme and the Health Insurance Marketplace, according to the joint HHS announcement from earlier this year.
In that February rollout, HHS Secretary Robert F. Kennedy Jr, Vice President JD Vance and CMS Administrator Mehmet Oz framed the shift as moving from a decades-old approach of paying claims first and investigating later to what the agency calls a “detect and deploy” model, using AI tools to flag suspicious claims at the point of adjudication.
The numbers behind the push are large enough to explain the urgency. Medicare’s fee-for-service programme alone made an estimated $28.83bn in improper payments in fiscal 2025, according to a CMS fact sheet; Medicare Part C added another $23.67bn.
A separate Government Accountability Office report in April put government-wide improper payments at roughly $186bn for the year, with the bulk concentrated in five programmes, including Medicare and Medicaid.
The regulatory vehicle behind the initiative is a formal Request for Information that HHS and CMS opened in late February, soliciting industry input on analytics methodologies, AI tooling and data-sharing approaches.
The RFI closed on 30 March and feeds into a planned proposed rule that CMS has been calling CRUSH, for “Comprehensive Regulations to Uncover Suspicious Healthcare”.
The May initiative appears to be the operational step that follows from that consultation, although neither HHS nor CMS has yet published the full vendor list or technical architecture behind it.
Pilots have been running in parallel. The HHS Office of Inspector General has tested a machine-learning model that scores providers for billing behaviour statistically associated with fraud and abuse, and CMS reported that total Medicare programme-integrity savings rose 59% in fiscal 2025, from $26.3bn the previous year to $41.9bn.
The agency attributes part of that jump to enhanced screening of new enrolees, including a six-month nationwide moratorium on new home health and hospice enrolments that took effect on 13 May.
The substantive risk in moving from post-payment review to pre-payment AI screening is what false positives do to providers. A flagged claim that delays payment to a legitimate practice, particularly a small one, is a material liquidity event. Industry groups have already pressed CMS, through the RFI process, for clear appeal rights and human review thresholds before any AI-flagged denial becomes final. None of those guardrails have yet been written into rule.
What HHS has not disclosed: which model vendors are being used, whether the system will operate on de-identified or fully identifiable claims data, and how the agency will audit the models’ own error rates.
The CRUSH rulemaking is the document those answers will eventually have to live in. For now, the initiative goes live against a backdrop of unusually large improper-payment numbers and a federal appetite for AI in compliance that is, by recent standards, unusually high.
