New Records Obtained by EFF Expose More Problems with Medicare's AI Experiment

New records obtained through EFF’s lawsuit reveal problems in Medicare’s WISeR program, which uses AI to assess prior authorization requests for selected medical services. CMS launched the program only six months after announcing it, despite providers’ calls for delay. In its first three months, the agency sent requests for information or corrective-action plans to at least four of the six vendors over delays, technical failures, poor communication, and possible noncompliant decisions. CMS questioned Virtix about slow determinations and apparently inaccurate denials, raised concerns about Zyter’s late notices and provider support, and cited Innovaccer for failing to deliver a working portal. Vendors said the compressed launch contributed to problems; one had warned CMS it might not be ready. The records include more than 800 pages, among them applications from 66 companies and vendor contracts.

The documents also leave key safeguards difficult to assess: some information is redacted, and claims of clinician review, monitoring, and audit trails do not show whether protections prevent improper denials. Human reviewers may defer to AI recommendations, especially if they lack time or authority to challenge them. Vendors also cite data such as patient demographics, provider behavior, and historical claims, but disclose too little to judge whether those inputs fairly reflect individual medical circumstances. EFF argues that outsourcing consequential decisions to private AI vendors, particularly under a payment model it says may reward denials, threatens patients’ access to care. It calls for greater transparency, oversight, testing, and congressional limits on automated decisions in Medicare.