A new study finds Medicare's AI prior authorization pilot has already cut common treatments by up to 20 percent across six states. The government calls it a war on waste. The researchers who measured it cannot yet say whether patients lost care they actually needed.
By Blumefield | October 29, 2026
The first real evidence
For most of its history, traditional Medicare paid its bills with remarkably few questions asked. That era is ending. On January 1 the federal government switched on a pilot that routes a growing list of treatments through AI prior authorization before it will pay for them, and this week brought the first hard evidence of what that does to patients.
Researchers at the University of Minnesota and Brown University reported on October 26 in JAMA that the program cut use of its targeted services sharply in its opening months. Epidural steroid injections fell 15 percent in the six pilot states compared with the rest of the country. Other flagged procedures dropped by roughly 20 percent. For an agency that has spent years warning about wasteful spending, those are the numbers it wanted to see.
The trouble is what the numbers cannot tell anyone. The researchers tracked how often procedures happened, not whether the people who skipped them ended up better or worse. In the carefully worded conclusion of a study that CMS will surely cite as vindication, the authors said the central question remains open. That gap, between fewer procedures and better medicine, is the whole argument over AI prior authorization in miniature.
What Washington switched on
The program carries a bureaucratic name that gives away its purpose: the Wasteful and Inappropriate Service Reduction Model, or WISeR. It is run by the CMS Innovation Center, the agency's in-house laboratory for testing new ways to pay for care, and it is scheduled to run for six years, from January 1, 2026 through the end of 2031.
WISeR applies only to traditional, fee-for-service Medicare, the half of the program that private insurers do not administer. That is a meaningful line. Prior authorization has long been routine in Medicare Advantage, where it has drawn lawsuits and congressional scrutiny, but traditional Medicare mostly stayed clear of it. WISeR imports AI prior authorization into the government's own book of business for the first time at scale, and it does so in six states chosen to stand in for the country: Arizona, New Jersey, Ohio, Oklahoma, Texas and Washington.
The list of services under review reads like a catalogue of expensive, discretionary medicine. It covers skin and tissue substitutes for chronic wounds, implanted electrical nerve stimulators, cervical spine fusion, certain knee procedures and the epidural steroid injections the new study singled out. CMS frames all of it as care prone to overuse, and it leans on its longstanding estimate that waste accounts for as much as 25 percent of US health spending. The full design is laid out on the agency's own WISeR Model page.
Who gets paid when care is denied
What makes WISeR different from ordinary red tape is who sits on the other side of the desk. CMS hired technology companies, one per state and including Cohere Health, Zyter and Virtix Health, to run the AI prior authorization reviews. Those firms do not simply shuffle paperwork. Under the model's design they earn a share of the money the government saves when they head off care judged unnecessary.
That payment structure is the pressure point. An analysis by the nonpartisan health research group KFF noted that the vendors are rewarded based partly on the volume of care they deny, an arrangement critics say builds a financial reason to say no. CMS has tried to wall that incentive off. It says no request can be rejected without sign-off from a qualified human clinician, that the companies face audits, and that their pay is adjusted on measures such as provider experience rather than denial counts alone. Whether those guardrails hold under a model that pays for savings is exactly what the next five years will test.
The dollars explain why CMS is willing to try. Skin substitutes alone cost traditional Medicare about $10.3 billion in 2024, the bulk of all spending on WISeR services, and the average price of a single skin-substitute treatment climbed from roughly $2,300 in 2019 to $21,200 in 2024. That is an 820 percent jump in five years for a category of product with thin evidence behind much of its use. When prices move like that, the case for a gatekeeper writes itself.
A 20 percent cut, and a question no one can answer
The JAMA study is the first real measure of whether AI prior authorization is actually closing that gate. Using Epic's Cosmos database, which draws on more than 300 million patient records, the researchers compared monthly use of targeted services in the six WISeR states against every other state and the District of Columbia, over the two years before launch and the first six months after. The design cannot prove cause and effect, but the pattern was clear enough.
It was also fading. The drop in epidural steroid injections shrank from about 21 percent in the first quarter of 2026 to about 9 percent in the second. The decline in other targeted services eased from 21 percent to 18 percent. Procedures that were not subject to the new reviews barely moved, which points to WISeR, rather than some broader shift in medicine, as the cause. Providers may simply be learning how to clear the new hurdles, or steering patients toward treatments that avoid them.
None of that settles the only question that matters to a patient. "The key question we can't yet answer is whether the care not received was wasteful or necessary," said lead author Hannah Neprash. Her Brown co-author, Michael Barnett, put the cost-benefit test bluntly: "We need to watch closely to see if the additional work of prior authorization justifies relatively little change in who gets care." The study measured volume, not pain relief, complications or denials, and it captured only six months of a six-year experiment.
Delays, politics, and a six-year bet
Away from the data, the friction is already visible. A report from Senator Maria Cantwell found that UW Medicine was waiting 15 to 20 days for decisions that used to take one to three, the kind of delay that can matter a great deal when a patient is in pain. In response to that pushback CMS has started letting providers with strong approval records earn exemptions from the reviews, a release valve borrowed from private insurance. At an October 15 conference a senior CMS official conceded that many parts of the program were not yet working as intended.
The politics have been just as rough. The Senate voted 46 to 50 in July against a resolution to kill the AI prior authorization pilot, leaving it alive but contested, and Democrats including Representative Suzan DelBene have called it a threat to patients. An earlier attempt to defund any traditional-Medicare prior authorization model passed a House committee in 2025 but did not survive into the spending law signed in February. For now, WISeR stands.
What is really at stake is bigger than six states. The Innovation Center designs its pilots precisely so it can scale the ones that work, and CMS has left the door open to expanding WISeR nationwide if the findings hold. A 20 percent cut in targeted procedures, with no proof yet of harm, is the sort of result that could justify exactly that. The question the JAMA authors could not answer is now one the rest of the country has a direct stake in. Before AI prior authorization becomes the default setting for American medicine, someone will have to show that the care it blocks is care nobody needed. For more on how automated systems are reshaping health and public policy, visit Blumefield.