Medicare's AI Prior Authorization Pilot Delayed and Denied Care Across Six States
What happened
Prior authorization is the mechanism Medicare uses to confirm a procedure or prescription is medically necessary before it gets paid for. It already slows care. The Centers for Medicare and Medicaid Services chose to speed it up by automating part of the review with an AI-assisted system called WISeR, running a pilot across six US states. The data that came back did not show a faster process. It showed a system generating widespread delays, technical failures, and denial rates that patients and providers found difficult to challenge.
The WISeR pilot routed prior authorization requests through AI evaluation before any human reviewer touched them. In the states where it ran, providers submitted requests into a system that was frequently slow to respond, prone to technical errors, and configured to deny a significant share of what came through. Patients waiting on approvals for surgical procedures, treatments, or durable medical equipment found their care stuck in a queue with no clear path to resolution and no obvious mechanism to escalate when the system produced the wrong answer.
Federal records, which the Electronic Frontier Foundation obtained and released, documented the human cost in concrete terms. Patients reported pain while waiting for procedures the system had delayed or blocked. Some surgeries were canceled outright. Others described emotional distress from fighting a process they did not understand and could not easily appeal. These were not rare edge cases in a pilot that otherwise worked. They appeared as a pattern across the record, consistent enough that the EFF's release framed them as systemic rather than incidental.
The structural problem here runs deeper than a bad implementation. Prior authorization is not an administrative inconvenience. It is the checkpoint between a physician's clinical judgment and a patient actually receiving care. Handing that checkpoint to an automated system, without adequate human oversight or a clear escalation path for errors, turns a coverage question into a medical delay. High denial rates from an AI system are not a calibration problem to tune out over time. They are policy outcomes, and the patients on the receiving end of them did not volunteer for a pilot that treated their procedures as test cases.
What the EFF records reveal is that the evidence of harm existed inside federal systems and was not visible to the public until someone pulled it out through records requests. That gap, between what an automated decision system does and what is accessible to oversight bodies, patient advocates, and the people it directly affects, is exactly the accountability gap that documentation infrastructure is designed to close. A provable record of what a system decided, when it decided it, and what the downstream effects were should not require a records battle to produce. When it does, the harm has usually already landed.
Reported impact
- Affected parties
- Not publicly disclosed
- Harm type
- Not publicly disclosed
- Scale
- Not publicly disclosed
- Financial impact
- Not publicly disclosed
- Regulatory action
- Not publicly disclosed
Classification
Relevant governance controls
Governance control mapping is not available for this record.
- No controls mapped
Not publicly disclosed
Control mapping is analytical. It does not state that any control would have prevented the incident.
Sources and evidence
This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.