State & Local GovernmentIssue 412 records
State Government
CMS said Oct 15 it will soon publish results of its Indiana Medicaid AI prepayment-fraud pilot with Oracle, is running pilots with five more states, and wants a national platform, while admitting parts of its AI-supported WISeR model are not working. No pilot accuracy data has been released. Minnesota's Medicaid office is asking CMS to disclose the criteria and any machine learning or AI behind federal flags on 870 providers before it acts. The edition's one pattern: states are being asked to act on federal AI fraud flags without the method or evidence behind them.
- Evidence records
- 2
- Cross-source patterns
- 1
- Evidence classes
- 2 independent reporting
- Outcomes
- 1 emerging1 cautionary
- Source freshness
- 2 new this fortnight
- Research completed
- 2026-10-19
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
State Medicaid agencies are being asked to act on federal AI fraud flags without seeing how the flags are made
Interpretation. CMS says it is piloting AI prepayment fraud detection with Indiana and five more states and wants a national platform, yet no accuracy or false-positive results have been published, and CMS concedes its AI-supported WISeR model has problems. Separately, Minnesota says CMS flagged 870 providers using analytics that include machine learning and AI, and the state is asking for the criteria and method before it acts. Together they show federal AI detection scaling faster than the evidence and review process states need to take defensible action.
Operating questionBefore your agency holds a payment or acts against a provider on an externally generated AI flag, what method, evidence and appeal path do you require, and who signs off?
Supporting evidenceCenters for Medicare & Medicaid Services (Deputy Administrator Kimberly Brandt); Indiana Family and Social Services Administration; OracleMinnesota Department of Human Services (temporary Medicaid Commissioner John Connolly); Centers for Medicare & Medicaid Services
Evidence ledger
Full record · every source keeps its link and limitations
CMS (Oct 15) says it will soon publish results of its Indiana Medicaid AI prepayment-fraud pilot with Oracle and is running pilots with five more states, while conceding parts of its WISeR AI prior-authorization model are not working
Speaking Oct 15 at the ADAPT conference, CMS Deputy Administrator Kimberly Brandt said CMS will soon publish results of the Indiana Medicaid fraud pilot it ran with Oracle and is actively working with five other states on Medicaid pilots. Per her slideshow, the pilots use AI to flag upcoding, phantom services and geographic anomalies in near real time before payment, and test a shared case platform for CMS, Medicaid Fraud Control Units, program integrity contractors, states and law enforcement; the stated goal, if proven, is a national standardized platform. No pilot results were reported. Brandt also said 'There are a lot of things about WISeR that are not working,' referring to CMS's AI-supported Medicare prior-authorization model, and the slideshow lists lessons: continuous oversight, essential human clinical review, and provider feedback as a monitoring signal. Why now: Indiana's pilot was announced June 24 as a 90-day effort, so its proof-of-concept report is due, and CMS says it intends to scale the model to more state Medicaid programs.
Why it matters, evidence and limitations
- Why it matters
- State Medicaid agencies own claims systems, payment holds and provider enrollment, so a federal prepayment AI platform would run on state data and change state workflows. Five additional states are already in pilots, and others may be asked to join. The WISeR admissions are the first official acknowledgment in this cycle that an AI-supported federal health program model has operating problems.
- Evidence and measured results
- Source findings (reported from Brandt's remarks and slideshow): Indiana pilot with Oracle complete or near-complete with results 'soon'; five other states in active pilots; AI flags suspicious billing before payment; shared investigation platform under test. Operator claims not attributable to the AI pilots: $41.9 billion FY2025 savings (59% increase) from the broader anti-fraud push; Medicaid War Room stopped $342 million in improper payments and sent 124 notices of intent to exclude as of September 2026; about 67% of watchlisted new providers and suppliers cut. Enrichment from Indiana FSSA's June 24 release: objectives include prepayment detection, recommended claim edits and prior-authorization triggers, AI-generated investigation-ready case packages, a scalability assessment and a final proof-of-concept report; WISH-TV (June 24) reported a 90-day term, removal of personal and protected health information before sharing with Oracle, and that the tool cannot initiate regulatory action. No accuracy, false-positive, baseline or sample data are available.
- Limitations and uncertainty
- Reporting of conference remarks and a slideshow; the slideshow was not inspected. No pilot outcomes, accuracy or false-positive data were released. The five other states are unnamed. Savings figures describe the broader anti-fraud program, not AI. Indiana pilot design comes from the June 24 state release and June WISH-TV reporting, not from results.
Minnesota's Medicaid office asks CMS to disclose the criteria, methodology and any machine learning or AI behind federal flags on 870 providers before the state acts on them
StateScoop reviewed a letter from Minnesota's temporary Medicaid Commissioner John Connolly to CMS Administrator Mehmet Oz. CMS identified 870 Minnesota providers for fraud or aberrant billing using historical billing information and CMS data analytics, which federal officials described as including machine learning and AI capable of analyzing Medicaid claims. Minnesota asks CMS to disclose the criteria, methodology and any AI or fraud-detection analytics used, saying it needs the underlying evidence before it can withhold payments or act. CMS has threatened to withhold $2 billion a year and has frozen $550 million in deferrals. Why now: on Oct 15 CMS said it will publish Indiana AI pilot results, is piloting with five more states and aims for a national prepayment platform, which makes Minnesota's unanswered request a live question for every state that receives federal AI flags.
Why it matters, evidence and limitations
- Why it matters
- This is a state agency stating, in writing, that it cannot act on AI-generated risk designations without the method and evidence behind them. Any state that receives federal or vendor AI flags faces the same question about due process, auditability and who bears responsibility for wrongful actions.
- Evidence and measured results
- Source findings: 870 providers identified by CMS; federal description of the technology as including machine learning and AI; Minnesota's written request for criteria, methodology and analytics; $2 billion annual withholding threatened and $550 million in deferrals frozen, including $259 million in February 2026; state says it has met all Corrective Action Plan requirements. Minnesota officials' claim: CMS is blurring criminal fraud and administrative actions. Context: a June 2026 Minnesota law created a Human Services Systems Modernization Advisory Council. No information on how the 870 designations were produced or how many were confirmed.
- Limitations and uncertainty
- Single-source reporting of a letter that was reviewed but not published; the letter itself was not inspected. CMS's response is not reported. The dispute is politically charged, and the state's characterizations are advocacy. Published Oct 8, before the last completed run; included for its relevance to CMS's Oct 15 scaling plans.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.