News · 2026-09-26

Records reveal delays in Medicare's AI-supported WISeR review program

Newly released records show serious delays and thousands of denials in WISeR, Medicare's AI-supported prior-authorization experiment in six states. The development matters because the program combines automation with human clinical review and pays participating companies partly from avoided spending, making its operational safeguards and incentives unusually consequential for patients. CMS says no AI alone makes a non-payment recommendation, but the records raise questions about whether the surrounding workflow is safe, timely, and accountable.

Key facts

• WISeR began in January 2026 and runs in six states through 2031.
• EFF says two vendors denied 5,944 prior-authorization requests in the first three months.
• One request was reportedly unanswered for 83 days despite a 72-hour target.
• Primary source: EFF's report on records obtained through FOIA litigation.

WISeR stands for Wasteful and Inappropriate Service Reduction. It is an Original Medicare model, not Medicare Advantage, that requires prior authorization or pre-payment review for selected services. The official CMS program page says participating technology companies may use enhanced technologies, including artificial intelligence and machine learning, to support review. It also says licensed clinicians make all recommendations for non-payment and that the program does not change Medicare coverage policy.

That distinction is necessary but not sufficient. “AI denied care” is a misleading shorthand if it implies a machine autonomously made every final decision. But an AI-supported system can still change who gets reviewed, what information is prioritized, how quickly a request is handled, and how a clinician encounters a case. A human signature at the last stage does not make earlier automation irrelevant. Think of airport security: a human can be the person who sends a bag for secondary inspection, while the scanner and queueing system still determine which bags reach that person and how long travelers wait.

EFF's report draws on about 1,000 pages of CMS records released through FOIA litigation. It says two vendors denied 5,944 requests in WISeR's first three months; that Virtix denied more requests than it approved; and that Innovaccer temporarily auto-approved requests because its system was not ready. The most vivid service-quality figure is an 83-day unanswered request against a stated 72-hour target. Provider records describe patients waiting in severe pain. These are operational reports from released records, not proof that every denied procedure was medically appropriate or inappropriate.

The payment model sharpens the accountability issue. CMS says participants are paid a percentage of expenditures associated with averted wasteful or inappropriate care, adjusted by performance measures. That is a familiar value-based design: a reviewer earns money if it helps avoid spending that should not occur. It can encourage waste reduction, but it also creates a tension between denying or delaying care and measuring patient harm accurately. The newly released records make that tension more than theoretical.

EFF's central criticism is transparency. It says training data, bias testing, hallucination testing, and monitoring for the vendors' systems remain largely undisclosed. The program's public description tells readers that clinicians are involved, but does not provide an independently auditable picture of how a model affects triage, what errors it makes for different populations, or how rapidly an erroneous denial is reversed. Those are the questions that determine whether human oversight is meaningful rather than ceremonial.

There is also a legal wrinkle. The Government Accountability Office ruled in May that the WISeR notice is a rule subject to Congressional Review Act submission requirements because it imposes new provider obligations and can affect beneficiary claim determinations. The GAO decision does not hold every WISeR decision unlawful, nor does it decide whether the use of AI is inherently improper. It identifies a procedural vulnerability in the way the policy was issued.

CMS's counterargument deserves full weight. The agency is not presenting WISeR as an unsupervised algorithmic gatekeeper. It says the model is voluntary, uses clinical review, applies only to defined services, includes quality measures, and seeks to reduce unnecessary or inappropriate care without changing coverage rules. Prior authorization already exists in healthcare, and some form of more efficient review could spare patients unnecessary interventions and reduce waste. The records alone cannot establish a causal link from a particular model output to each delay or denial.

The stronger critique is structural, not rhetorical. When an automated system is embedded in a high-stakes administrative pipeline, the right audit is not simply “was a human somewhere in the loop?” It is: what did the system recommend, how often did humans disagree, how fast did cases move, who bore the burden of an error, and how did financial incentives affect behavior? The 5,944-denial figure and 83-day delay make those questions urgent because they turn abstract model-governance concerns into consequences for people waiting for care.

WISeR is a useful test of a broader policy claim: AI can make bureaucracy more efficient without making it less humane. The evidence now public does not settle that claim, but it means the program should be judged with detailed outcome reporting rather than marketing language about “human oversight.” In medicine, a system that is technically accurate but routinely late can still fail patients.

Primary source, verified: read the paper →

Ground Truth. (2026, September 26). Records reveal delays in Medicare's AI-supported WISeR review program. Ground Truth. https://groundtruth.day/news/medicare-wiser-ai-prior-authorization-records.html

@misc{groundtruth:medicare-wiser-ai-prior-authorization-records,
  title  = {Records reveal delays in Medicare's AI-supported WISeR review program},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {sep},
  url    = {https://groundtruth.day/news/medicare-wiser-ai-prior-authorization-records.html}
}

Topics: policy · healthcare · ai-governance · accountability · cms