New records raise questions over Medicare’s AI prior-authorisation programme

New records published by the Electronic Frontier Foundation (EFF) show that US officials raised concerns with several vendors in a Medicare programme using AI to assess prior-authorisation requests. The documents add detail to a consequential question: what oversight is actually catching problems when automated systems help shape access to medical care?

Watch Desk analysis

What happened

EFF says it has published more than 800 pages obtained through a lawsuit, including applications from 66 companies, contracts and correspondence between the Centers for Medicare & Medicaid Services (CMS) and vendors. The records include requests for information or corrective action plans sent to at least four of the programme’s six vendors during its first three months.

In its account of the documents, EFF says CMS raised concerns including delayed determinations, possible inaccurate or inconsistent denials, poor communication and technical problems. It names Virtix, Zyter and Innovaccer among the vendors discussed. The vendors’ responses are also included in the released records; EFF says they describe fixes and, in one case, earlier warnings about readiness.

Why it matters

The WISeR programme uses AI to evaluate prior-authorisation requests for some Medicare services. A delay or mistaken denial can matter to a patient waiting for care, so the practical question is not simply whether a clinician is nominally in the loop. It is whether that review is timely, informed and able to challenge an automated recommendation.

The records do not settle that question. EFF says information about some safeguards was redacted and that the public still lacks key details about how vendors use data and how well their safeguards work. Its claims about the programme’s problems and implications are the organisation’s interpretation of the documents, not a finding that every vendor or decision failed.

Our read

The useful development is the paper trail: CMS was querying multiple vendors about problems, and those vendors’ responses are part of the record. That gives patients, clinicians and policymakers something more concrete to examine than a reassuring label such as “human review”. But without fuller information about the safeguards and how decisions are checked, the documents leave the central accountability question open.

What to watch

• Whether CMS or the vendors provide fuller details about the concerns and their responses.
• What information about safeguards remains redacted, and whether it is later released.
• Whether the programme publishes evidence on review quality, delays and disputed decisions.

Discussion spark: What should count as meaningful human oversight of an AI-assisted medical decision: a clinician reviewing every denial, or evidence that reviewers have time and authority to challenge the system?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.