Prior authorization was originally designed with a sensible goal: before a health insurer pays for an expensive treatment, a doctor has to justify it. Done well, it catches unnecessary procedures and controls costs. Done badly, it delays care until patients give up.
AI has made the second version faster and cheaper to run at scale.
A survey of 1,000 U.S. physicians by the American Medical Association found that 93% say prior authorization delays care, and 82% say it causes patients to abandon treatment altogether. Those numbers have not changed in years. What has changed is speed: insurers can now process denials in bulk, with little or no human review, using automated systems.
The most striking evidence is in the courtroom. A class action lawsuit against UnitedHealth, one of the largest health insurers in the world, centers on an AI tool called nH Predict. According to the plaintiffs, this tool's denial decisions were reversed on appeal more than 90% of the time. The problem is the other side of that statistic: only about 0.2% of policyholders ever filed an appeal. So the algorithm was losing almost every fight it picked, aimed at people who almost never showed up to fight.
In March 2026, a federal court ordered UnitedHealth to open its internal records showing how the tool works. Cigna faces a similar suit over its own automated review system. Humana's lawsuit survived a dismissal attempt in August 2025. These are not isolated incidents; they represent a pattern across the largest names in health insurance.
The picture gets sharper when you look at Medicare Advantage, the version of U.S. public health insurance run by private companies. A Congressional report found that denial rates at UnitedHealthcare, CVS, and Humana's plans rose significantly as their AI use increased. One Minnesota health system nearly stopped accepting UnitedHealthcare patients entirely because denial rates were reportedly ten times higher than competing insurers.
At the same time, the U.S. federal government is pushing AI deeper into the system. A new Medicare pilot program called WISeR launched in January 2026 across six states, using AI to screen prior authorization requests for 17 types of services. The program runs until 2031. CMS says final denials must be reviewed by a licensed clinician, and that vendors are paid to get decisions right, not to deny. Critics, including hospital groups and several members of Congress, are not convinced. The vendors' pay is reportedly tied to savings generated, which creates pressure toward denial.
For any business operator with employees on health plans, this matters directly. The AMA survey found that 58% of physicians treating working adults said prior authorization had already affected those patients' job performance. Delays in care mean delays in returning to work. And doctors spend an average of 13 hours per week just on paperwork related to these approvals, with 40% employing staff dedicated entirely to that task. Those costs flow through to everyone.
Some states are starting to act. Arizona, Maryland, Nebraska, and Texas all passed laws in 2025 that ban insurers from using AI as the sole reason for a denial. Colorado is rolling out rules requiring appeal rights for AI-generated decisions. But federal law remains weak, and Congress has so far failed to pass a bipartisan reform bill that had support from both parties.
The most useful thing to know: if you or an employee receives a denial, appeal it. Across all of Medicare Advantage in 2024, more than 80% of appeals were overturned. Most people never try. AI systems, at least as currently deployed, appear to be built on the assumption that they won't.