AI and Prior Authorization: Faster Denials, or Fewer?
Anyone who has tried to get a health insurer to approve a treatment their doctor actually recommended will not need this explained to them. The prior authorization process, in theory, exists to stop unnecessary procedures and control costs. In practice, it has become a bureaucratic obstacle course that delays care, exhausts patients, and burns through physician time. Now AI is being offered as the solution. Whether it will fix anything, or just automate the dysfunction more efficiently, is genuinely unclear.
The numbers are grim enough without adding algorithms to the mix. A 2025 Commonwealth Fund survey found that around one in five working-age American adults with private insurance reported being denied coverage for doctor-recommended care this year. Of those, 41 percent said the denial delayed their treatment, and more than a quarter said their condition got worse while they waited. These are not edge cases.
AI proponents argue that the technology could at least speed up approvals for straightforward claims, cutting through paperwork that currently sits in queues for days. That is a reasonable argument on its face. The problem is who is deploying it and why. A 2025 American Medical Association survey found that 61 percent of physicians are worried AI will make denial rates worse, not better. Health policy analyst Camm Epstein put it plainly: AI should be used to make appropriate care easier to approve, not necessary care easier to deny. Sensible framing, routinely ignored.
The Trump administration has launched a pilot programme called WISeR, which stands for Wasteful and Inappropriate Service Reduction Model, running in six states through 2031 under the Centers for Medicare and Medicaid Services. It uses machine learning alongside human clinical review to scrutinise procedures CMS considers vulnerable to overuse or fraud, including knee arthroscopies, nerve stimulator implants, and certain skin and tissue treatments. Prior authorization has rarely been used in original Medicare before, so this is a meaningful expansion.
The incentive structure deserves scrutiny. Vendors operating within WISeR get a cut of what CMS calls averted expenditures. In plain terms, they earn more when care is denied. That is not a subtle conflict of interest. Early reporting from the Washington Post, KFF Health News, and the Seattle Times suggests the model has already produced care delays and denials across its pilot states in its opening months. Several legislators have moved to block its funding.
Meanwhile, on the private insurance side, CMS Administrator Mehmet Oz has taken the opposite tone, effectively telling insurance executives to ease up on prior authorization or face regulation. The industry has responded with data showing an 11 percent drop in prior authorization requests between mid-2025 and early 2026. What the denial rate is doing, nobody is saying.
Insurers have also pledged that AI is not being used to deny claims without human clinical review, and that they will be more transparent about the reasoning behind rejections. Whether that holds up to scrutiny is another question. The AMA has been pushing for exactly this kind of transparency, along with detailed clinical justifications for denials, for some time.
Physician and Healthcare Huddle founder Jared Dashevsky captured the frustration well. AI could, in theory, cut administrative waste and free up clinical time. Instead, what is being built looks more like an arms race, faster denials, faster appeals, more automation layered onto a system that was already broken. That assessment is hard to argue with when the financial incentives for the companies involved point so clearly toward rejection rather than approval.
The underlying problem with prior authorization is not really a technical one. It is a structural one. Adding machine learning to a process designed to obstruct care does not reform the process. It just makes the obstruction harder to challenge.