Medicare WISeR: How AI Prior Authorization Works (2026)

Short Answer

WISeR is a Medicare pilot in which six technology companies review requests for certain medical services in six states before Medicare pays, starting January 1, 2026 and running through December 31, 2031. The Centers for Medicare & Medicaid Services (CMS) says all recommendations not to pay are determined by "appropriately licensed clinicians." The companies are paid "a percentage of the expenditures associated with averted wasteful, inappropriate care," which critics point to as an incentive to say no. CMS's page does not say who issues the final denial and does not describe appeal rights, so those are the two questions to take to the agency before assuming anything.

This page explains what the model is, where it runs, who is paid and how, what is and is not established, and what to do if you are a Medicare patient in one of the six states. We wrote it from CMS's own model page and from the reporting linked below. It is general information, not medical or legal advice.

What WISeR is

WISeR is a model run by the CMS Innovation Center. It adds a prior-authorization step, meaning a review before payment, to traditional Medicare for selected items and services. Adding that step is a real change for patients and practices in the pilot states. CMS says the model "will run for six performance years from January 1, 2026 to December 31, 2031."

CMS's page gives only examples of the services covered: skin and tissue substitutes, implantation of electrical nerve stimulators, and knee arthroscopy for knee osteoarthritis. Industry coverage describes seventeen categories in all, but the CMS page we read does not list them, so look at the model's current service list if a particular procedure matters to you.

Who runs it, and where

CMS lists six participating vendors, each tied to a state and a Medicare Administrative Contractor region:

Trade coverage notes this is the first CMS Innovation Center model in which the main participants are technology companies rather than providers or insurers. That framing is accurate for who is at the table, and it is why the model is discussed as an AI story: the vendors are described as using AI-enabled review technology. What exactly the software decides, and what a person decides, is not spelled out on the CMS page.

How the vendors are paid

CMS states that vendors receive "a percentage of the expenditures associated with averted wasteful, inappropriate care," and that the percentage is adjusted for performance measures that include provider experience. That is the detail that drives most criticism. When a reviewer's pay rises with the amount of care that does not get paid for, the incentive points toward denial, even if every individual decision is made in good faith. CMS's counter is the clinician requirement and the performance adjustments.

It is worth separating three claims people often blend:

• Established by CMS's own page: the dates, the states, the vendors, the payment structure, and the statement that non-payment recommendations are determined by licensed clinicians.
• Reported by others: that the vendors use AI-enabled review, that there are seventeen service categories, and that House Democrats introduced legislation to repeal the model. We read these in news and trade coverage, not in a statute or a CMS document.
• Not established here: how many requests have been denied, how many appeals succeeded, or whether patients were harmed. We found no figure we could source, so we are not reporting one.

What the Electronic Frontier Foundation reported

The Electronic Frontier Foundation published an article in October 2026 titled "New Records Obtained by EFF Expose More Problems with Medicare's AI Experiment." Its section headings include "WISeR's Rollout Was Rushed and Rocky" and "More Transparency Needed About AI Safeguards." We link to it rather than summarize its findings, because we have not independently reviewed the records it describes. If you are researching WISeR, read the original and weigh it as an advocacy organization's account.

If you are a Medicare patient in a WISeR state

We cannot tell you whether a specific service needs review, because the model's service list and rules can change. These steps are general and cost nothing:

• Ask your provider's office whether a requested service is part of WISeR. Practices in the six states are the ones dealing with the requests day to day.
• Ask for the decision in writing. If a service is not approved, ask who made the determination and what the reason was.
• Ask how to appeal. CMS's model page, as we read it, does not describe appeal rights, so get the process from 1-800-MEDICARE or your state health insurance assistance program before a deadline passes.
• Keep records. Note dates, names and what was said when you called.

For a broader look at how AI shows up in health coverage decisions, see our pages on AI lawsuits and AI legislation.

The question to watch

The design question is not whether a model uses AI. It is whether anyone outside the vendor can see how the system decides, check its error rate and correct a wrong denial quickly. CMS has said licensed clinicians determine non-payment, and the pilot is scheduled for six years. Over that time, data on denial rates, appeal reversals and delays would be the evidence that settles the argument. As of this writing, we have not found published figures of that kind that we could source.

Questions worth putting to CMS and the vendors

Because the public record is thin, the most useful thing a journalist, lawmaker or patient advocate can do is ask specific questions. These are the ones we would ask, and each can be answered with a number or a document:

• How many requests did each vendor review in each state, and how many were approved, not approved or pended for more information?
• Of the non-payment recommendations, how many were later reversed on appeal, and how long did the process take?
• What exactly does the software do, and at which step does a licensed clinician look at the case? Is that clinician employed by the vendor?
• How is the performance adjustment to the vendor's percentage calculated, and does it include patient outcomes or only spending?
• What protections exist if a delayed approval postpones care a patient needs?
• Will CMS publish results before the six-year model ends, and who will evaluate them?

A plain reading of the model is that it tests whether a payment-linked review can reduce spending on services CMS considers low value. Whether it also protects patients is a different question, and only data on delays and reversals can answer it.

Sources and method

This is an analysis of public documents and reporting. We did not test any vendor's system and have no access to its decisions.

• CMS, WISeR Model (dates, states, vendors, payment structure, clinician statement)
• TechTarget, Meet the 6 vendors participating in the CMS WISeR Model (vendor roles and service categories)
• STAT News, Medicare WISeR prior authorization pilot tech vendors
• Electronic Frontier Foundation, New Records Obtained by EFF Expose More Problems with Medicare's AI Experiment