Records obtained by the Electronic Frontier Foundation (EFF) through a lawsuit against the U.S. government have revealed operational problems, delays, and denials of care requests within the Wasteful and Inappropriate Service Reduction (WISeR) model, a Medicare pilot program that uses artificial intelligence to help evaluate prior authorization requests for certain medical services.
The EFF published approximately 1,000 pages of documents it received from the Centers for Medicare & Medicaid Services (CMS), including contracts with technology companies, internal status reports, and notes from care providers. The organization says the records reflect earlier complaints about delayed approvals, inappropriate denials, system failures, and the impact on patients.
Delays Exceeding the Announced Deadline
CMS launched the WISeR model in January 2026, six months after announcing it, and applied it to older adults in six states. Under the model, providers must request prior authorization for certain treatments if they want to ensure Medicare coverage, while private companies contracted by CMS evaluate the requests using artificial intelligence tools.
CMS says the companies must respond within 72 hours, but internal status reports obtained by the EFF show that a large number of requests did not receive a response within that timeframe. One document mentions a request that remained unanswered for 83 days. Care providers also reported that patients waited weeks for approvals or instructions, leading to postponed medical procedures and leaving patients in pain.
Financial Incentive Linked to Denials
The documents say that the contractors are paid for requests they deny and receive no payment when a denial is overturned on appeal. CMS ties payments to quality scores measuring the accuracy and speed of decisions, but the WISeR data-reporting guide indicates that lower scores reduce payments by only 5% to 10%, according to the EFF.
During the program’s first three months, two companies denied more than 20,000 prior authorization requests. Virtix denied more requests than it approved during the same period, and CMS asked the company to submit a corrective action plan.
Rushed Launch and Technical Problems
The records show that Innovaccer informed CMS about a month before the launch that it would go live with a version lacking full functionality and that had not undergone complete testing. The company pointed to changing requirements, unclear governance mechanisms, and insufficient time for comprehensive testing with care providers. It said that enabling automatic approval for all requests was the available option because CMS had decided not to postpone the model’s start date.
Later reports indicate that some functions had still not been completed or tested several months after launch. Innovaccer handled requests in Ohio, where care providers’ notes included complaints about poor communication and long waits; statements said that patients waited more than a month for an application number to be issued, resulting in canceled surgeries.
Why Does This Matter?
The documents provide a practical indication that using artificial intelligence in medical coverage decisions is not solely about model accuracy, but also about contract design, launch speed, and appeal and oversight mechanisms. CMS says that a qualified physician must review every denial, but the EFF points out that fundamental questions about the systems used and tests for accuracy, bias, and hallucinations remain without complete answers.
This issue is becoming more important because a planning document from June 2025 discussed expanding the model in the future to cover 13 services in its first year, with the possibility of adding more urgent services such as air ambulance transport, cancer treatment, MRI scans, and drugs whose coverage criteria are not publicly available. CMS continues to provide additional documents in the case, so the full picture of the systems and safeguards used is still being completed.