CMS's AI Medicare Denial Pilot Rejected Over Half of One Vendor's Requests, GAO Finds Procedural Violations

The Centers for Medicare and Medicaid Services launched a pilot program in January called WISeR, short for Wasteful and Inappropriate Service Reduction, that uses artificial intelligence and machine learning to decide whether Medicare will pay for certain procedures before a patient gets them. It's running in six states, New Jersey, Ohio, Oklahoma, Texas, Arizona, and Washington, and is scheduled to continue through the end of 2031.

Traditional Medicare has never required this kind of prior authorization before. WISeR applies it to roughly a dozen services, including nerve stimulation, epidural steroid injections for pain, cervical spinal fusion, skin substitutes for wounds, and treatments for incontinence and impotence.

The stated goal, according to CMS's own program description, is protecting taxpayers by cutting fraud, waste, and abuse in Medicare spending. That's a legitimate objective. Nobody serious disputes that Medicare fraud costs taxpayers real money, and using technology to flag suspicious billing patterns isn't inherently a bad idea.

The execution is where things fell apart.

The Numbers Don't Lie

Documents obtained by the Electronic Frontier Foundation through litigation, and first reported in detail by Ars Technica, show one contracted vendor, Virtix, denied more requests than it approved. A weekly report dated March 30 showed Virtix had reviewed 6,096 prior authorization requests, approving 2,863 and denying 3,233, a 53% denial rate. A separate analysis published by AI Governance confirmed that figure and noted several vendors missed the program's mandated 72-hour decision window entirely.

The delays weren't minor. The EFF documents show at least one request sat pending for 83 days. One healthcare provider surveyed by CMS said a surgery got pushed back almost two months because of what they described as zero communication from the system. Another provider called the program "a disgrace to the human race" and reported watching patients cry in pain while waiting on decisions.

Another vendor, Zyter, reportedly couldn't tell the difference between Medicare Part A, which covers hospital care, and Part B, which covers outpatient services, for months, causing data discrepancies. Zyter CEO Sundar Subramanian told Ars Technica in a written statement that the company is now "fully functional across Medicare Part A and Part B claims" and is working with federal partners to improve the process.

A third vendor, Innovaccer, was so unprepared for the January rollout that it asked CMS to delay it. When CMS said no, Innovaccer temporarily set its system to auto-approve every request just to avoid a total backlog, writing to officials that "auto-affirming is the only path available that avoids creating a backlog of unprocessed prior authorizations and claims while we finalize, validate, and deploy the full rules-based solution."

GAO Found a Procedural Problem

In May, the Government Accountability Office determined that CMS did not follow proper administrative procedure when it set WISeR up, which raises real legal questions about the pilot's footing. No court has ruled the program illegal, and no criminal investigation has been opened. But the procedural finding is exactly the kind of thing that tends to end up in front of a judge if litigation continues.

Despite that, CMS has moved forward with the pilot and has outlined plans to expand it in coming years, according to Ars Technica's reporting.

Congress Split on Party Lines

Rep. Suzan DelBene (D-Wash.) called a committee vote last week to force the administration to release additional WISeR documents. Republicans on the committee voted it down. DelBene said in a statement that "it's clear why the administration is doing everything they can to conceal these documents," adding that "the more that Americans learn about WISeR, the more outraged they get at the Trump administration for jeopardizing their care and trying to privatize Medicare."

That's DelBene's characterization. Republicans blocking the document release didn't offer a public counter-explanation in the reporting reviewed here for why more disclosure was rejected. Seniors who paid into Medicare for decades are the ones stuck waiting on an algorithm that, by CMS's own contractor's data, denied more than half of one company's caseload.

The Bigger AI Picture

The WISeR mess is unfolding against a backdrop where the administration has pushed back hard against AI regulation generally. President Trump told reporters in Ireland on September 13 that AI critics represent "very negative forces" bringing up "things that won't happen," and Pentagon technology chief Emil Michael has argued existing product liability law and executive authority are sufficient oversight, telling Newsmax's Carl Higbie that new congressional AI rules would "turn us into Europe."

Whatever the merits of that stance for frontier AI models, WISeR is a live example of what happens when a federal agency deploys AI in a high-stakes decision system without adequate testing or oversight. It didn't take a rogue superintelligence to cause harm here. It took an unprepared vendor, a missed deadline, and a senior with an 83-day wait for a decision on their care.

The open question is whether GAO's procedural finding leads anywhere. No court has blocked WISeR, no funding has been cut, and CMS's own timeline runs through 2031. Whether that changes depends on pending EFF litigation and whether Congress, controlled currently by Republicans, decides to force more transparency before the midterms.