Medicare's AI Authorization Experiment

Bill Curry drives five hours round trip to get an epidural injection for his back. He's done it for years. This year, Medicare's new AI-powered prior authorization system added a third trip — just to fill out a form describing how he feels. He hasn't gone. He's thinking about driving to Kansas instead.

That's the human cost of deploying AI at scale, on an accelerated timeline, without adequate testing — and then insulating it from accountability through the complexity of government contracting. KFF Health News reports on the rollout of WISeR, the Wasteful and Inappropriate Service Reduction Model, a Trump administration program using AI to manage prior authorization for 13 categories of medical services in six pilot states. The program launched in January 2026, roughly seven months after it was announced. Government contractors were still adding features in the spring.

The early results are not good.

What's Happening in the Pilot States

The University of Washington's medical system had nearly 100 patients waiting for epidural injections due to WISeR-related delays as of this spring. A New Jersey patient waiting on back pain authorization ended up in the hospital — more expensive care, precisely the outcome the program claims to prevent. An Oklahoma radiologist documented payment delays of six to eight weeks for claims that are supposed to be resolved in 15 days. A Phoenix pain management doctor hadn't been paid for nine epidurals as of May.

Doctors in the pilot states believe AI hallucinations are driving some of the denials. One Arizona physician described a rejection stating his patient wasn't eligible for procedures in the thoracic region — the patient needed an injection to the neck. An Oklahoma radiologist documented four times that a patient lacked numbness; the denial cited numbness as the reason for rejection. The vendor's CEO said he hadn't heard of any AI hallucinations. The doctors filing appeals tell a different story.

CMS says AI approves applications quickly when clinical data supports approval, with an "immediate yes" in 88% of cases. That figure sounds reassuring until you consider that the 12% remainder, applied across Medicare's scale, represents an enormous number of patients navigating delays, appeals, and denials generated by a system that vendors describe as human-reviewed but practitioners describe as AI-driven.

The Accountability Gap Is the Story

The structure here is worth examining. Private contractors build and operate the AI systems. Humans are said to make final decisions. When errors occur, CMS says it "appreciates feedback" and will use it to "help providers better understand WISeR processes." Nobody is named. Nobody is responsible. The vendor whose product generated the hallucination says he hasn't heard about it.

This pattern shows up wherever AI is deployed at an institutional scale without clear accountability mechanisms. The technology handles volume that humans can't. The errors that fall outside its parameters land on patients with no recourse except the appeals process — which, in this case, is funded by the same government that pays the contractors to run the system that generated the appeals in the first place.

A University of Pittsburgh health policy researcher quoted in the KFF piece puts it directly: prior authorization may simply shift costs to patients and doctors rather than save them. The government pays contractors to handle appeals it didn't budget for because of a system it rolled out faster than the evidence warranted.

What This Means Beyond Healthcare

The WISeR program is a healthcare story, but the failure mode it illustrates belongs to every industry deploying AI at scale right now. Speed without adequate testing. Vendor accountability insulated by contractual complexity. Real-world harm is absorbed by the people with the least power to push back. A 12% error rate that sounds manageable in a spreadsheet and looks catastrophic when it's a Medicare patient in rural Oklahoma who can't get an epidural approved.

For marketers and growth leaders building AI into decision-making workflows, the lesson here is not that AI doesn't work. It's that AI deployed without clear accountability structures, human oversight that's more than nominal, and genuine feedback loops will generate harm proportional to its scale. The faster you move, the more expensive those errors get.

The AI programs worth building are the ones that answer the question: when this makes a mistake, who is responsible, and how does the person affected get it fixed? If that question doesn't have a clear answer in your workflow, the WISeR program is an instructive preview of what insufficient guardrails look like when they meet the real world at volume.

The AI strategy that holds up over time isn't the fastest one. It's the one that earns the trust of the people it affects.

Winsome Marketing helps growth teams build AI programs with real accountability and human judgment built in from the start. Let's talk about what that looks like.