Medicare's AI Prior-Authorization Pilot Pays Vendors to Deny. The FOIA Records Show the Cost.

Medicare's new AI prior-authorization pilot pays vendors each time they deny a coverage request. Internal records released this month through an Electronic Frontier Foundation FOIA lawsuit show what that incentive bought in practice: requests sitting unanswered for 83 days, contractors blowing past CMS's own 72-hour response window, and at least one vendor warning the agency weeks before launch that the system was not ready.

Prior authorization is the process a doctor has to clear before an insurer will pay for a treatment. It is the gate that decides whether a procedure happens, and Medicare has historically run it through its contractors, not algorithms. The WISeR model (Wasteful and Inappropriate Service Reduction), launched in January 2026 across six states, is the first time traditional Medicare has handed that gate to AI vendors, and the first time it has paid those vendors on the back end of the decision.

Under the WISeR model, contractors are paid for requests they deny, but not for denials that get reversed on appeal. That means a vendor's revenue scales with how aggressively it cuts coverage, while the cost of getting an appeal right falls on the doctor and the patient. It is the difference between paying a reviewer to be accurate and paying a reviewer to be a gate.

The FOIA records, roughly 1,000 pages of contracts, status reports, and provider complaints published by EFF in September, show how the design played out in the first months. One internal report flagged a prior authorization request that went unanswered for 83 days. Timeliness data for two of the WISeR vendors in January 2026 show a significant share of requests missed CMS's stated 72-hour turnaround. Those are not edge cases: the WISeR Model FAQ treats 72-hour responses as the expected baseline, and the FOIA records show the contractors missing it during the first reporting period of a six-state pilot that federal officials have flagged as a possible template for national expansion.

A vendor did try to slow the rollout. According to the FOIA-released records, one WISeR contractor told CMS weeks before the January launch that its system was not ready, then went ahead under what the document describes as CMS's decision not to delay. The agency has not publicly addressed that warning. STAT+ reported this week on the same document set; the EFF records and an earlier Washington Post investigation from March 2026 (as summarized by the Center for Medicare Advocacy) are the public source of the vendor names, complaint excerpts, and timeliness tables STAT's recap sits on top of.

Ohio doctors have told reporters they may stop offering certain complex treatments because the prior-authorization burden has gotten worse, not better, under the AI pilot. Approval rates for WISeR-reviewed procedures are reported to be lower than in Medicare Advantage, the private-plan version of the program that already uses AI in coverage decisions. The Center for Medicare Advocacy's review of the early data calls the rollout "troubling." That is not the same as a national finding: the pilot has been live only since January, in six states (Arizona, Ohio, Oklahoma, Texas, New Jersey, and Washington), and covers roughly a dozen procedures. The case for or against AI in traditional Medicare has to be made on those six states, not on a hypothetical national program.

A federal program that pays reviewers per denial, ignores a contractor's pre-launch warning, and then misses its own response deadlines is the proof of concept now sitting in front of CMS. The agency has said the WISeR model could be expanded in the future. The next review point is whether CMS publishes the contractor-by-contractor timeliness data the FOIA records contain, and whether it restructures the payment model before the pilot grows beyond the six states it covers today.