About a thousand pages of CMS records concerning Medicare’s AI prior authorization pilot were released in mid-September through a Freedom of Information Act lawsuit brought by the Electronic Frontier Foundation. They are worth reading closely if you sell artificial intelligence into healthcare, and not for the reason you might expect.
The Wasteful and Inappropriate Service Reduction model, known as WISeR, launched in January 2026 across Arizona, New Jersey, Ohio, Oklahoma, Texas and Washington. It uses private vendors applying AI to review prior authorization requests for roughly a dozen services, including epidural steroid injections and skin substitutes.
The records describe requests resolved well past the program’s own 72 hour standard, including one reported instance that went unanswered for 83 days. They also describe a vendor notifying CMS ahead of launch that a working product was not realistic on the agency’s timeline, and indicating it would auto-affirm requests until development was complete. Reporting on the records indicates that feature development remained incomplete months later.
The incentive structure is the element that should concern anyone building in this space. Vendors retain a share of the savings generated on services they decline, while the reported penalty for a poor quality score reduces payment by roughly 5 to 10 percent.
This matters commercially even for companies with no connection to utilization management. Every health system executive and every payer medical officer now has a federal reference case for artificial intelligence making coverage decisions badly. That reference case will be present in the room during the next procurement conversation, whether or not anyone names it.
The companies that get through it will be the ones able to describe their error rate, their human review pathway, and their appeal reversal data without being asked twice. Trust in this market is becoming a documented artifact rather than a posture.