The Medicare program is using AI-assisted technology to review requests for certain medical services for about 6.4 million Original Medicare beneficiaries across six US states, according to a Congressional Research Service briefing.

Papers released after a freedom of information lawsuit show high denial rates and waits well beyond the three-day federal target. The Wasteful and Inappropriate Service Reduction model, known as WISeR, launched on 1 January 2026, with participant portals accepting requests from 5 January for services provided from 15 January.

Documents Show High Denial Rates Across the Six-State Pilot

The Centers for Medicare and Medicaid Services says WISeR uses artificial intelligence and machine learning, with a human clinician required to review every refusal.

The trial covers Arizona, New Jersey, Ohio, Oklahoma, Texas and Washington and runs until 31 December 2031. Six technology firms hold the state contracts, including Virtix Health in Washington and Innovaccer in Ohio.

First-year services include skin substitutes, epidural steroid injections, cervical fusion, knee arthroscopy and several nerve stimulators. Providers may seek prior authorisation or have the claim undergo pre-payment medical review.

The Electronic Frontier Foundation sued CMS over access to the records and has since published about 1,000 pages obtained through the litigation. Those files show that two vendors alone denied 5,944 prior-authorisation requests during the first three months.

Virtix Health refused more requests than it approved in that period. CMS later required the firm to submit a corrective action plan.

EFF's filing cited Texas data showing that about 62 per cent of requests were approved initially, rising to 84 per cent after human review. CMS says participants are paid a share of savings only when they cut wasteful care.

Medicare AI Prior Authorisation Left Some Files Waiting Weeks

CMS requires WISeR participants to issue decisions within 72 hours for requests submitted through their electronic portals, although requests routed through Medicare Administrative Contractors can take additional time.

Status reports record thousands of requests that missed that mark. Several hundred were still unanswered at the end of March, including one that had sat for 83 days.

Tori Noble, a staff attorney at the foundation, said: 'The public has a right to know more about the algorithms driving decisions around their healthcare.'

Senator Patty Murray, a Washington Democrat, wrote on X: 'AI has no business denying doctor-recommended procedures.'

Providers in Ohio complained about Innovaccer, which had warned before launch that it needed more time. In June CMS required Virtix Health to set out root causes after early waits of about five days.

Key facts for patients covered in the six states remain tightly drawn. Only listed procedures need advance review. Emergency and inpatient-only care sit outside the model. Appeal rights are unchanged. A clinician, not software alone, must sign every non-affirmation.

Vendors Paid From Savings as Federal Oversight Tightens

CMS says the model tests technology-assisted review processes similar to those already used by other payers, including Medicare Advantage plans. Participants earn money from averted spending on requests that fail existing criteria and are not later overturned.

EFF has said the payment structure can give vendors a financial incentive to identify care as wasteful, while CMS says payments are subject to performance and quality safeguards. A gold-card exemption for high-compliance doctors is planned during 2026. The Senate rejected a July bid to halt the pilot.

CMS is still producing files on Medicare AI prior authorisation as vendors work through backlogs in the six states.

Officials say poor performers face corrective action or removal from the model, and that existing coverage rules have not been rewritten.