Medicare's New AI Pays to Deny

Introduction

"Model payments are calculated from requests that did not result in a paid claim." That's a direct quote from the CMS Provider Factsheet for the new Medicare AI prior authorization program. The American Hospital Association and the Electronic Frontier Foundation both say the number attached to that sentence is up to 20%.

What CMS Built, and What They Called It

The program is called WISeR, short for Wasteful and Inappropriate Service Reduction. It's a Medicare prior authorization model run by CMS's Innovation Center, launched January 1, 2026, and scheduled to run through December 31, 2031. It operates in six states with six contractors covering 14 service categories in Performance Year 1, including cervical fusion, epidural steroid injections, hypoglossal nerve stimulation for sleep apnea, and skin substitutes. All of this sits under Section 1115A of the Social Security Act, which matters for reasons that come back later.

The part worth staring at is how the contractors get paid. According to the Federal Register notice that launched the program, "model participants are compensated based on a share of averted expenditures." The Provider Factsheet puts it in plainer English: payments "are calculated from requests that did not result in a paid claim." The AHA, in its October 2025 comment letter to CMS, translates: "CMS has indicated that participating vendors will be compensated by receiving 10-20% of the savings associated with care denials." AHA called it "a perverse incentive to deny care."

The federal government built a Medicare AI whose revenue model is the volume of treatment it blocks. It launched the program under CMMI demonstration authority that bypassed standard notice-and-comment rulemaking. And it set up the legal architecture so the class-action remedy currently live against UnitedHealth doesn't apply here.

Where the 20% Comes From

The number shows up in a filed federal pleading first. The EFF federal complaint, filed March 25, 2026, Case 3:26-cv-02591 in the Northern District of California, says in paragraph 13 that WISeR vendors "are compensated, in part, on the volume of healthcare services they deny; they are entitled to as much as 20 percent of the associated savings."

The AHA letter puts a floor under it. "10-20%" is the range the hospital industry's own comment letter cites after being briefed by CMS. Neither CMS's public factsheet nor its FAQ names a specific percentage. That's the gap the EFF FOIA lawsuit is trying to close: the actual contract terms, accuracy testing, bias evaluations, and audit records. EFF filed its original FOIA request January 29, 2026. As of the March 25 complaint, CMS had released zero records in response, not even partial or redacted documents.

Then the Texas data. From the same EFF complaint, paragraph 15, citing Washington Post reporting from March 17, 2026: in Texas, 62% of WISeR prior authorization requests were initially approved by the AI. After a human clinician reviewed the AI's output, 84% were approved. The filing compares that to Medicare Advantage's national approval rate of "more than 92 percent." That comparison is not apples-to-apples, because WISeR's 62% covers only 14 specifically selected high-waste service categories while the 92%+ is an all-service national MA average. The 22-point jump between the AI's first call and the human's final call is the number I keep circling back to, because it's the volume of denials the algorithm produces before a clinician overrides it.

On the physician side, the 2024 AMA Prior Authorization Physician Survey, released February 2025, has the clinical backdrop. 94% of physicians say prior authorization results in somewhat or significantly worse patient clinical outcomes. 29% report prior auth has caused a serious adverse event for a patient in their care, which AMA defines as hospitalization, permanent impairment, or death. Those figures describe the prior authorization system overall, and WISeR joins it with a percentage-of-denials incentive attached.

And the six contractors are not strangers. According to the CMS Provider Factsheet, the awarded vendors are Cohere Health (Texas), Genzeon (New Jersey), Humata Health (Oklahoma), Innovaccer (Ohio), Virtix Health (Washington), and Zyter (Arizona). All six come from the Medicare Advantage prior authorization space. The Federal Register notice explicitly states CMS selected vendors with "demonstrated success in managing prior authorization processes with enhanced technology for other payers/health plans." The same tooling they run for private MA plans, now running on Medicare dollars.

The Safeguards, Honestly

CMS did build real guardrails, and the piece doesn't work if you pretend they don't exist. Every denial requires a human clinician to sign off, so the AI cannot deny a claim on its own. Contractors get financially penalized for inaccurate determinations, and successful appeals trigger recoupment of payments already disbursed. Providers can resubmit without limit. And the CMS FAQ confirms gold carding is planned for mid-2026, which would exempt providers who hit a 90% provisional affirmation rate from further WISeR review.

The targeted services also aren't random. KFF's February 2026 analysis found skin substitute spending in traditional Medicare grew nearly 2,000% from 2019 to 2024, reaching $10.3 billion and accounting for 83% of all WISeR-covered service spending. Average price per skin substitute service jumped 820% in five years, and HHS OIG has repeatedly flagged the category as a fraud vector with documented abusive billing patterns.

KFF also notes CMS simultaneously implemented a separate nationwide skin substitute payment reform starting January 1, 2026, cutting Medicare payments for that category by roughly 90%. That single policy change addresses the biggest spending target WISeR was built for. KFF's own conclusion is that WISeR's first-year fiscal impact is "likely to be modest." A straightforward price cut may already be doing the work WISeR's denial incentive was built to do.

Who Gets Paid

The six contractors are the direct beneficiaries. AHA's letter confirms 10-20% of averted expenditures flows to them. WISeR-covered services totaled $12.3 billion of Part B spending in 2024 per KFF, so even a modest reduction scaled across six performance years is meaningful contractor revenue on top of the federal contract itself.

CMS also gets a spending reduction it can label "waste, fraud, and abuse" rather than "Medicare cut." And CMS used Section 1115A CMMI demonstration authority to launch the program, which let it skip the Section 1871 notice-and-comment rulemaking that a standard Medicare rule requires. Senators Wyden and Gillibrand, in their September 18, 2025 letter to CMS Administrator Mehmet Oz and CMMI Director Abe Sutton, wrote that by "classifying the model as voluntary, CMMI circumvented the traditional notice and comment rulemaking process. This approach allowed CMS to fast track the model implementation without public input." They also pointed out the model is "functionally mandatory" because providers who don't participate in prior auth face mandatory pre-payment medical review instead.

Medicare Advantage plans get a quieter benefit. If Original Medicare adopts prior authorization, one of Original Medicare's structural advantages over MA evaporates. The pressure on MA plans to reform their own prior auth practices lessens because the federal program now mirrors theirs. AHA flagged that exact concern in its letter.

The Sovereign Workaround

The legal piece is what separates WISeR from every private insurer story. UnitedHealth is currently defending the Lokken class action over its NaviHealth subsidiary's nH Predict algorithm, with plaintiffs alleging a 90% error rate on appealed denials. A federal judge allowed breach-of-contract and breach-of-implied-covenant claims to proceed in February 2025. We covered the nH Predict case in detail when the court ordered UnitedHealth to hand over internal algorithm documents by April 29 — including cost-savings projections — specifically to test whether the tool was built as a clinical instrument or a revenue mechanism. That case exists because Medicare Advantage is delivered through private insurance contracts governed by state law, which creates a cause of action patients can actually bring.

WISeR patients don't have that door. Section 1315a(d)(2) of the Social Security Act explicitly bars judicial and administrative review of CMMI demonstration decisions. Beneficiaries retain standard Medicare administrative appeals and can eventually reach federal court under 42 U.S.C. § 405(g), but those are individual claim appeals inside the Medicare administrative pipeline. The state-law breach-of-contract and bad-faith class action currently proceeding against UnitedHealth is not available against the federal sovereign under the same legal theories.

The Morgan Lewis healthcare regulatory group published an analysis in July 2025 calling WISeR contractors potential "modern-day bounty hunters" and comparing the structure to CMS's earlier Recovery Audit Contractor program, which faced sustained provider criticism for aggressive denial behavior. AHA's letter explicitly compared the WISeR compensation model to MultiPlan, whose percentage-of-savings structure is the subject of a federal antitrust lawsuit and a New York Times investigation. Both comparisons come from inside the industry, not from patient advocates.

What Happens Next

The incentive is in the government's own factsheet: "Model payments are calculated from requests that did not result in a paid claim." Human clinician review is supposed to contain the AI's bias toward denial. Performance penalties are supposed to correct bad actors. The separate skin substitute price fix is supposed to defang the biggest fraud target without the denial incentive mattering much. And the legal structure shielding CMMI demonstrations is supposed to hold if cases start coming. The Texas 22-point gap between AI approval and post-human-review approval is the first real-world data point testing any of that.

Two things to watch. The EFF FOIA lawsuit is the mechanism that could force the actual contract percentages, accuracy data, bias testing, and audit records into public view. As of early April 2026, CMS has produced nothing, and the case is live in federal court. The Seniors Deserve SMARTER Care Act, introduced November 7, 2025 and sitting in the House Energy and Commerce Committee, would prohibit WISeR and any substantially similar model. A House Appropriations amendment to block WISeR funding cleared committee in September 2025 but got stripped from the Consolidated Appropriations Act signed in February 2026, so the votes to kill the program aren't in place yet. The open question is whether the first public denial data CMS actually releases will change that math.