Documents Show Medicare's AI Prior Authorization Pilot Was Rushed Into Launch

For someone waiting on approval for a pain injection or a wound care product, the difference between a smooth review process and a broken one isn't abstract. It's the difference between getting treatment on time and living with pain, or a wound, a little longer than necessary. That's the real-world stake behind a Medicare pilot program that newly released documents suggest was pushed out the door before it was ready.

The program is called the Wasteful and Inappropriate Service Reduction model, or WISeR for short. It uses artificial intelligence to help decide whether Medicare will approve certain procedures and products before a patient receives them, a process known as prior authorization. Think of it like a gatekeeper checking your ticket before you enter a concert, except here the gatekeeper is software, and the ticket is your doctor's recommended treatment.

WISeR launched in January and currently touches six states: New Jersey, Ohio, Oklahoma, Texas, Arizona, and Washington. It targets specific services, including skin substitutes used in wound care and epidural injections for pain management. The Centers for Medicare and Medicaid Services designed the pilot to run through 2031, giving it years to prove whether AI-assisted review can actually reduce wasteful spending without harming patients along the way.

That's the promise, anyway. More than a thousand pages of documents and data, obtained by the Electronic Frontier Foundation through a Freedom of Information Act lawsuit, tell a messier story about how the program actually came together.

A Vendor's Warning, Apparently Ignored

Among the most striking revelations is that one of WISeR's vendors told CMS directly that the agency's timeline was unrealistic. The vendor warned officials it would not be able to deliver a working product by the date the federal agency wanted to launch the program. CMS moved forward regardless.

This kind of warning matters because prior authorization systems, especially ones powered by AI, are not simple on-off switches. They require careful calibration to distinguish between genuinely unnecessary care and treatment a patient actually needs. Rush that calibration, and the risk isn't just technical glitches. It's real people experiencing delayed or denied care because the system underneath the decision wasn't fully baked.

The documents obtained by EFF reportedly detail additional errors beyond the timeline warning, though the full scope remains behind a subscription paywall on the original reporting. What's already public is enough to raise concern: a federal health program, covering millions of older Americans and people with disabilities, moved ahead with an AI tool despite an explicit heads-up that it wasn't ready.

This is not a hypothetical concern about AI in health care. Prior authorization already carries a difficult reputation among patients and providers alike, often criticized for delaying necessary treatment even under traditional review processes. Layering AI onto that system without adequate testing time doesn't just risk repeating old problems. It risks compounding them, potentially at greater speed and scale, since automated systems can process and deny claims far faster than human reviewers ever could.

Regulatory oversight exists precisely to catch these kinds of mismatches between ambition and readiness. When an agency receives a direct warning from its own vendor and proceeds anyway, it raises legitimate questions about who is accountable when the predictable happens. Patients in the affected states didn't get a vote on the launch date. They're simply subject to whatever the system decides, whether it's calibrated correctly or not.

It's worth remembering why CMS created WISeR in the first place. The stated goal is reducing wasteful and inappropriate services, a legitimate policy aim given how much unnecessary spending exists in the American health care system. Nobody disputes that some prior authorization requests genuinely involve overuse or misuse of Medicare resources. The question isn't whether the goal is worthwhile. It's whether the execution matched the stakes involved.

Why It Matters

Six states and years of runtime mean this pilot's problems won't stay contained to a single news cycle. Every day the program operates under a rushed foundation is another day patients in New Jersey, Ohio, Oklahoma, Texas, Arizona, and Washington depend on a system that at least one vendor said wasn't ready for prime time. Skin substitute treatments and epidural injections aren't optional luxuries. They're tools clinicians use to manage chronic wounds and persistent pain, conditions that don't improve while paperwork sits in an algorithmic queue.

There's also a broader lesson here for how federal agencies adopt AI more generally. WISeR is one of the higher-profile examples of AI directly influencing whether Medicare beneficiaries receive care, and it's now also one of the clearer examples of what happens when deployment timelines are set by bureaucratic convenience rather than technical reality. If CMS wants AI-assisted prior authorization to work, and to earn public trust, it needs a development process that treats vendor warnings as decision points, not obstacles to route around.

The EFF's lawsuit that forced these documents into public view underscores something important too: transparency about how these systems get built is not optional. Patients, providers, and taxpayers all have a stake in understanding whether the tools making coverage decisions were built with the care those decisions deserve. As WISeR continues running through 2031, that scrutiny should only intensify, not fade. The people waiting on approvals right now are counting on it.