CMS aims to quadruple provider exclusions with anti-fraud AI in Medicaid: the challenge is not repeating WISeR

By ExecutiveGov · October 16, 2026.

The U.S. Centers for Medicare & Medicaid Services (CMS) will soon publish the results of an artificial intelligence pilot in Indiana designed to detect Medicaid fraud before payments go out. Deputy Administrator Kimberly Brandt announced this at the ADAPT conference, as reported by ExecutiveGov citing Nextgov/FCW. The pilot was carried out with Oracle, and CMS is working with five other states on similar pilots.

According to Brandt's presentation, these systems use machine learning to detect "upcoding" (billing for a more expensive service than the one provided), charges for services never delivered and anomalous geographic patterns in near real time. They are also testing a shared workspace where CMS, states, law enforcement and fraud control units pursue the highest-risk cases together. If it works, CMS intends to roll it out nationwide.

The figures CMS provides are the agency's own and have not been independently audited in the material: program integrity efforts produced $41.9 billion in savings in fiscal year 2025, up 59% from the previous year. The "Medicaid War Room," launched in April, had blocked $342 million in improper payments and issued 124 exclusion notices as of September. Brandt also cited a "Netflix-style" algorithm that compares new providers with existing ones and monitors suspicious ones for six months; according to her, about 67% of those flagged have been removed for noncompliance.

The second front is legal: the Department of Health and Human Services (HHS) this week gave CMS the authority to bar providers from federal health programs, including Medicare, Medicaid and TRICARE. Brandt, who had lobbied Congress for that power, expects suspensions to quadruple within six months.

The most interesting part is the self-criticism. Brandt acknowledged that "there are a lot of things about WISeR that aren't working." WISeR is a six-year Medicare pilot of AI-assisted prior authorization. According to the article, hospitals and Democratic lawmakers argue that it delays and denies care to older people; in June the House Ways and Means Committee tried to end it, but the Senate has rejected those attempts. The lessons CMS cites are continuous technology oversight, human clinical review and attention to provider feedback.

Our take: anti-fraud AI is one of the most defensible uses in the public sector, because it targets a real cost and its victims should be the fraudsters. But there are two asymmetries worth watching. The first concerns metrics: "savings" and "blocked payments" measure what gets stopped, not how many false positives are generated or how long it takes a legitimate provider to prove its innocence. Until the Indiana pilot publishes error rates and not just dollar amounts, the $41.9 billion figure is a headline, not an evaluation.

The second concerns power. An algorithm that flags, combined with an authority that can exclude and a forecast of quadrupling suspensions, creates incentives to optimize for volume. The WISeR case shows the danger: when AI decides on access to care, the patient pays the price. That is why human review is not window dressing; it is the condition for the tool to be legitimate.

More broadly, this is the hard short-term transition we defend without hiding it: the automation of government will arrive before its safeguards. In the long term, a health system that doesn't lose resources to fraud can devote them to care, and that is where AI adds value. Getting there requires measuring errors, publishing results and giving anyone wrongly flagged a fast route to appeal.