Prior authorization is the paperwork gate that stands between a patient and a procedure their doctor already wants to do. Private Medicare Advantage plans have used it for years, and it consistently ranks at the top of the list of administrative burdens that physicians hate, precisely because it turns "your doctor recommends this" into "please wait while someone you will never meet decides whether your doctor was right." Traditional Medicare, the government-run version most seniors picture when they think of Medicare, mostly did not work this way.

Then, on January 1, 2026, the Centers for Medicare & Medicaid Services (CMS) launched the Wasteful and Inappropriate Service Reduction Model, branded WISeR, and brought prior authorization, with an AI engine bolted on, into traditional Medicare for the first time.

How WISeR is supposed to work

WISeR is a six-year pilot running from 2026 through 2031 in six states: Arizona, New Jersey, Ohio, Oklahoma, Texas, and Washington. CMS estimates it touches around 6.4 million traditional Medicare beneficiaries. It applies to roughly a dozen services the agency considers "low-value" or vulnerable to overuse and fraud: epidural steroid injections, cervical spinal fusion, knee arthroscopy, nerve and spinal-cord stimulators, skin substitutes, and similar items. Emergencies and inpatient-only services are excluded.

Here is the part that matters for this site. CMS does not run the reviews itself. It contracts with private companies to use, in the agency's own words, "enhanced technologies, such as Artificial Intelligence (AI) and Machine Learning (ML)," alongside human clinical review, to decide whether each request gets approved. The official framing is that the model "protects American taxpayers" and steers patients toward evidence-supported care while easing administrative burden.

The framing did not survive contact with the rollout.

Pay-to-deny incentives

Before getting to the AI errors, it is worth sitting with the business model, because it shapes everything else. The private vendors running WISeR reviews can earn a share of the savings the program generates. Savings, in a prior-authorization program, come from not paying claims. A House summary of the program put it bluntly: CMS "contracts with private companies to use AI to review claims for certain procedures and pays those companies based on how many claims they deny."

You do not need a degree in incentive design to see the problem. When the entity deciding whether to approve your care gets paid more the more often it says no, "appropriate care" and "denied care" start to look uncomfortably similar from the vendor's side of the ledger. This is the same structural conflict that made prior authorization in Medicare Advantage so loathed, now imported into the program seniors chose specifically to avoid it.

When the denial cites symptoms the patient never had

The AI failures are where WISeR earns its headstone. Physicians and their staff who spoke with KFF Health News described denials that referenced clinical details that were garbled or simply made up. Reporters documented a case in Arizona where a denial stated a patient was ineligible for procedures in the thoracic region of the spine, when the patient actually needed an injection in the neck. In Oklahoma, a radiologist reportedly documented four separate times that a patient did not have numbness, only to receive a denial citing numbness as a reason.

These are not subtle judgment calls about whether a borderline procedure is warranted. They are denials premised on facts that contradict the patient's own chart. A model that "knows" a patient has numbness the chart explicitly rules out is doing the same thing IRCC's system did to a Canadian scientist and the same thing a chatbot does when it invents a citation: producing fluent, confident output untethered from the record in front of it. In a search box that is irritating. In a Medicare denial it means a real person in pain waits longer, or gives up.

The delays are measurable. In Washington state, patients reported waiting up to three weeks instead of days for an authorization decision, and far longer for the actual procedure. In an April 2026 letter drawing on hospital-association data, Senator Maria Cantwell wrote that "procedures that were typically completed within two weeks prior to WISeR now take four to eight weeks," forcing patients to reschedule repeatedly, "prolonging pain and allowing underlying conditions to worsen." The University of Washington's medical system alone had nearly 100 patients waiting earlier in the year for epidural injections because of WISeR-related delays. The Center for Medicare Advocacy summarized it with the kind of understatement that does a lot of work: the pilot "is not going smoothly anywhere."

Humans in the loop, allegedly

CMS and its vendors say humans make the final decisions on approvals. That is the same reassurance offered for nearly every consequential AI deployment, and it deserves the same scrutiny here as anywhere else. Doctors and their staffs told KFF Health News they believe AI is playing a large role in the process, and that some denials are the product of AI hallucinations that garble or fabricate information. When asked about the suspected AI-driven errors, a Medicare spokesperson said the agency appreciates "feedback on provider experience" that will help providers "better understand WISeR processes," which is a graceful way of changing the subject.

The structural issue is identical to the one in every "human in the loop" arrangement. If the system is designed to process high volumes quickly, the human review compresses toward a glance, and a glance does not catch a confident machine assertion that the patient has a symptom they do not have. The whole selling point of automating these reviews is speed, and speed is exactly what erodes the scrutiny that would catch the errors. When the human's job is to ratify the model fast, "final human decision" describes a signature, not an investigation.

A pilot that turned into a legal fight

WISeR did not just generate angry doctors; it generated a procedural problem for CMS. On May 12, 2026, the Government Accountability Office determined that the WISeR Model Notice is a "rule" under the Congressional Review Act, because it imposes new prior-authorization requirements that substantially affect the rights and obligations of providers and beneficiaries. In plain terms: CMS was supposed to submit it to Congress before it took effect, and it did not.

That finding opened a 60-day window for lawmakers to force a vote on repealing the model. On May 19 and 20, 2026, members of the House and Senate introduced companion Congressional Review Act resolutions to disapprove WISeR; in the House the resolution (H.J. Res. 187) was led by Representatives Greg Landsman and Suzan DelBene, with a Senate companion led by Senator Ron Wyden and others. CMS had already, in April 2026, delayed prior authorization for two of the listed services, deep brain stimulation and a type of lumbar decompression, citing "operational readiness." Operational readiness is a polite phrase for "this is not ready."

Why this belongs in the graveyard

It would be easy to file WISeR under ordinary policy controversy, the kind of thing reasonable people argue about. But the reason it earns a place here is specific and documented: an AI system, deployed in a safety-critical context by people paid to deny claims, produced denials citing clinical facts that were wrong or invented, and real patients absorbed the consequences in the form of delayed care and prolonged pain. That is not a debate about whether low-value procedures should be scrutinized. It is a deployment that failed at the one thing it had to get right, which is being accurate about the patient in front of it.

The honest version of cost control would put the burden of proof on the system, not the sick. If an AI flags a request, a clinician with time and authority should review the actual record before anyone is told no, and a denial that contradicts the chart should be impossible to issue, not merely appealable after weeks of waiting. Instead, WISeR built a fast, profit-motivated "no" machine and pointed it at seniors, then asked them to file appeals when it hallucinated. Saving Medicare money is a legitimate goal. Doing it by automating denials premised on symptoms the patient never reported is how you turn a budget line into a patient-safety story.