EFF publishes records exposing delays and questionable denials in Medicare’s AI pilot

Medicare officials challenged four contractors over delays, technical failures and other problems during the opening months of an AI-assisted approval pilot. More than 800 pages of records published October 27 by the Electronic Frontier Foundation expose those early failures—and raise questions about safeguards governing decisions on whether Medicare will pay for care.

EFF obtained the records through a lawsuit filed in March. The release includes federal requests for information and corrective action, vendors’ responses, applications from 66 companies, and contracts for implementing and evaluating the program. Together, they show what officials were questioning behind the scenes as providers struggled with the new process.

The Wasteful and Inappropriate Service Reduction model, or WISeR, began January 1, 2026, as a six-year pilot in six states. It uses private companies and AI to review selected services under traditional Medicare. Prior authorization means seeking approval for payment before treatment; the pilot aims to reduce waste and fraud.

Portals and decisions fell behind

The problems surfaced quickly. In the first three months, CMS—the Centers for Medicare & Medicaid Services—sent information requests or corrective-action requests to at least four of six vendors, according to EFF. The concerns extended beyond one company: contractors in Washington, New Jersey, Arizona and Ohio received warning letters.

A February 27 request to Zyter cited reports of late decision letters and inadequate processes for handling providers’ concerns. Innovaccer received two corrective-action requests and a noncompliance notice over an unfinished provider portal. Innovaccer told CMS it had warned months earlier that full portal functionality might not be ready at launch.

In Washington, CMS’s March 17 letter to Virtix Health listed slow determinations, portal gaps and poor customer service. It also questioned “denials that appear inaccurate or inconsistent with applicable criteria.” Those were concerns raised by the agency, not a finding that every rejected request was improper.

Records examined by KUOW show Virtix rejected 53% of 6,096 claims processed in the first quarter. Virtix attributed high rejection rates to incomplete provider documentation. In its earlier response to CMS, it also blamed processing bottlenecks and delayed guidance, saying more than 500 submissions were affected during the program’s first two weeks.

A confirmed improvement, with financial stakes

Virtix now says average processing takes 1.18 days for prior authorization and 1.17 days for reviews before payment, below the three-day requirement. CMS confirmed the turnaround improvement to KUOW. Virtix also says the approval rate reaches 84.3% after providers have an opportunity to resubmit requests with better documentation.

The payment structure adds another concern. Contractors receive 25% of the cost of procedures whose spending they avert, KUOW reported from the documents. Payments can be reduced by up to 10% for slow or inaccurate decisions. A federal memo acknowledged that the structure could incentivize denials of care.

Human review is required; its effectiveness remains unclear

CMS’s published FAQ requires a clinician with relevant expertise to review every non-affirmation—a request not approved during prior authorization. It also allows physicians to seek a discussion with a clinical reviewer when resubmitting a request. CMS says it will audit participants to prevent inappropriate denials, and patients retain appeal rights for denied claims.

EFF argues those requirements do not answer whether reviewers have enough time or practical authority to challenge AI recommendations. Some vendor descriptions of safeguards were redacted. Other applications listed human review, continuous monitoring and audit trails, but EFF says their effectiveness remains difficult for the public to assess.

The records also leave questions about data. Applications reference provider behavior, patient demographics, historical claims and performance metrics. EFF says the missing detail is how vendors use those inputs in individual decisions, making it difficult to assess whether broader patterns could introduce bias or outweigh a patient’s circumstances.

EFF is urging Congress to impose stronger legal limits on automated Medicare decisions. Its position goes beyond repairing portals or reducing backlogs: the organization argues that AI should not determine access to government benefits. The newly released records put specific vendor failures and unanswered oversight questions behind that demand.