CMS’s AI prior authorization pilot enters the public record under a cloud of delay and uncertainty
Medicare’s experiment with AI prior authorization was designed to do three things at once: catch unnecessary care, speed decisions, and hold down spending. Newly released records now suggest the program may have moved faster than its safeguards. The documentation, obtained after a Freedom of Information Act lawsuit, paints a picture of a federal pilot that was operational before key vendor systems were fully ready and before several implementation problems had been settled.
The model at the center of the controversy is the Wasteful and Inappropriate Services Reduction, or WISeR, initiative, which began on January 1, 2026. CMS enlisted six private technology companies to help review prior authorization requests for selected Medicare services in six states: New Jersey, Ohio, Oklahoma, Texas, Arizona, and Washington. The agency framed the effort as a way to protect taxpayers while pairing advanced technology with human clinical review. In practice, the released records show a program wrestling with software readiness, answer turnaround times, and the basic mechanics of appeal-like review.
One line from the records stands out: the launch appears to have outpaced the infrastructure meant to support it.
Vendor readiness problems surfaced before the model was fully underway
Among the first concerns documented in the records was whether participating vendors had systems ready for a live Medicare workflow. Innovaccer, one of the six companies involved, warned CMS that its provider portal might not be fully ready before launch. That kind of warning matters because prior authorization is not just a back-office task; it is the point at which a patient’s access to injections, devices, surgery, or other treatment can be slowed or redirected.
The records suggest the readiness issues were not confined to one company. A federal payment review system that depends on technology vendors has a low tolerance for rollout defects, because even small software failures can cascade into claim delays, unanswered requests, and confused providers. In a setting where prior authorization already frustrates clinicians, the addition of immature infrastructure risks turning a process intended to reduce waste into one that simply shifts burden downstream.
Denial rates raised questions about whether coverage rules were being applied consistently
CMS also scrutinized the performance of Virtix Health after the company rejected 53% of more than 6,000 prior authorization processing requests over a three-month period. A high denial rate does not automatically mean those decisions were wrong. Still, the figure drew attention because the company attributed the volume to problems processing an influx of claims, which raises a different concern: whether the system was operating at a scale and pace the vendor could not absorb without degrading accuracy.
Prior authorization rests on a narrow balance. It is supposed to prevent unnecessary services without blocking care that is clinically appropriate. If a vendor is denying large volumes because of processing failures, the policy goal of discouraging overuse starts to blur into a risk of inappropriate withholding. In Medicare, where many patients already face chronic disease and limited margin for delay, the stakes are not abstract. A denied or stalled request can mean no pain intervention, no sleep apnea treatment, no device, no surgery, and no timely relief from symptoms that worsen while the paperwork sits untouched.
An 83-day delay exposes how easily access can stall
One request sent to the vendor Genzeon went unanswered for 83 days. That is not a routine administrative slip. For a patient waiting on a procedure to reduce pain, restore mobility, or treat an escalating condition, nearly three months can become the difference between manageable symptoms and functional decline.
Prior authorization delays are often discussed as a paperwork problem, but the physiological reality is more direct. Pain can intensify, inflammation can persist, sleep can deteriorate, and mobility can deteriorate while a request sits in limbo. In procedures such as steroid injections, nerve stimulation, arthroscopic knee surgery, or sleep apnea treatment, time lost in authorization is time lost in treatment effect. A model built to streamline utilization review can backfire if the clock runs longer than the condition can tolerate.
The payment structure creates pressure that CMS will need to police closely
WISeR vendors can receive a percentage of expenditures tied to care deemed wasteful and then prevented through their reviews. CMS says compensation is adjusted based on accurate performance, but the structure still invites hard questions about incentives. A vendor paid in part for blocking spending has a financial interest that runs in the same direction as denial volume, even if accuracy is supposed to moderate that incentive.
That arrangement does not prove bias. It does, however, make transparency essential. If the financial reward favors savings, oversight has to show that the same system does not quietly reward over-denial. In prior authorization, incentives shape behavior quickly. A model that is intended to curb waste can become suspect if patients, clinicians, and regulators cannot see how denial decisions are validated, corrected, or reversed.
Human review may not be enough without visible safeguards
CMS requires a licensed clinician to review every denial recommendation. On paper, that creates a human check on automated judgment. The records leave unresolved how meaningful that review is in practice. If a clinician can only confirm an algorithmic recommendation with limited authority to override it, the “human-in-the-loop” safeguard may be weaker than it appears.
That problem is especially important in Medicare, where the consequences of a bad denial are not theoretical. Patients are older, often medically complex, and more vulnerable to delays than healthier populations navigating ordinary administrative processes. An AI-assisted workflow can help scale review, but scale alone is not a clinical virtue. The system has to prove that algorithmic screening is accurate, that escalation pathways work, and that human reviewers have real discretion when the facts of a case do not fit the model.
A federal test of AI in coverage review now faces a credibility test of its own
WISeR addresses a legitimate policy concern. Unnecessary procedures can expose patients to complications and add cost to Medicare. The theory behind the model is straightforward: use technology to triage low-value services faster, reserve clinical attention for the borderline cases, and reduce waste without harming access. The records now circulating around the program suggest the first months were consumed by implementation problems that should have been resolved before launch.
The bigger question is not whether prior authorization can be modernized. It is whether Medicare can outsource a critical access decision to private AI companies without sacrificing transparency, turnaround time, or accountability. The released documents offer early evidence that the program may be vulnerable on all three fronts. Congress will likely face pressure to examine whether the model’s legal and operational guardrails are strong enough to protect patients before the experiment expands any further.