In June, I published an analysis of the first six months of WISeR, the Medicare pilot that contracts with technology companies, including companies using artificial intelligence (AI), to review prior authorization requests. The public record at the time was limited to a corrective action plan for one vendor, a Government Accountability Office (GAO) opinion, and statements from provider groups and members of Congress.
On September 8, 2026, the Electronic Frontier Foundation (EFF) released approximately 1,000 pages of Centers for Medicare & Medicaid Services (CMS) records obtained through Freedom of Information Act (FOIA) litigation. This update summarizes what those records document, reports the status of each pilot state, and outlines the implications for providers, suppliers, and counsel, including those in states outside the pilot.
Background
The Wasteful and Inappropriate Service Reduction (WISeR) Model began January 1, 2026, under the CMS Innovation Center’s demonstration authority, and is scheduled to run through 2031. It requires prior authorization in Original Medicare (not Medicare Advantage) for a defined list of services, beginning with 13 categories that include skin substitutes, epidural steroid injections, and implanted nerve stimulators. Six states participate, each assigned one technology vendor. Vendor compensation is a percentage of the Medicare spending associated with services that are not approved, adjusted by quality measures. A provider may submit a claim without prior authorization, but the claim is then routed to prepayment medical review; in practice, participation is not optional. CMS states that a licensed clinician reviews every recommended denial.
Findings from the September records
Response times. CMS’s published standard is a vendor response within 72 hours. Internal status reports from the first months of the program show the standard was frequently missed, and one request remained unanswered for 83 days (released records, p. 234).
The quality-score adjustment. CMS has described vendor quality scores as the counterweight to a compensation model tied to reduced spending. The released data reporting guide shows that a low quality score reduces a vendor’s payment by 5 to 10 percent (p. 99). In my assessment, an adjustment of that size is unlikely to offset a financial incentive based on the value of services not approved. Of the findings in this release, I consider it the most significant for governance purposes, because it goes to whether the incentive structure was reasonably mitigated.
Launch readiness in Ohio. Approximately one month before launch, the Ohio vendor, Innovaccer, informed CMS that it would go live without full functionality or end-to-end testing and would approve all requests automatically in the interim. The vendor wrote that, given CMS’s decision not to delay the start date, automatic approval was “the only path available” (pp. 216–217). A status report from April 2026 shows features still incomplete (p. 200). Automatic approval spared Ohio beneficiaries from decisions made by an untested system. It also means Ohio’s early data cannot be used to assess the tool’s accuracy, and it establishes that CMS proceeded after its contractor reported that the system was not ready.
Reported effects on patients. A CMS-administered feedback survey of Ohio providers in March 2026 describes delayed procedures, waits of several weeks, cancelled surgeries, and, in one provider’s words, “patients calling our offices crying in pain.” The agency collected these accounts itself.
Services considered for expansion. A June 2025 planning document lists candidate services for later performance years, including services that would require urgent authorization: air ambulance transport, cancer treatment, and magnetic resonance imaging (MRI) (p. 22). This is a planning list. The record does not show that any of these services has been added. Some commentary has also listed cardiac procedures; EFF’s summary does not, and I have not included them pending direct review of page 22.
Correction to the reported denial count
EFF’s original post stated that two vendors “denied over 20,000” requests in the program’s first three months. The vendor-level figures do not support that number. My September 16 brief reconciled them as follows:
| State and vendor | Decisions through approximately March 30, 2026 | Denials |
|---|---|---|
| Washington (Virtix Health) | 6,096 (2,863 approved) | 3,233 (approximately 53%) |
| Arizona (Zyter) | approximately 14,300 | approximately 2,711 (approximately 19%; reported as “under 20%”) |
| Total | approximately 20,400 decisions | 5,944 denials |
Approximately 20,000 is the number of decisions, not denials. On September 15, EFF updated its post to state that the two vendors denied 5,944 requests. The accurate figure still supports a serious concern: a contractor compensated on reduced spending, and already under a corrective action plan for missed deadlines, denied more Medicare requests than it approved.
Status by state
| State | Vendor | Public record as of September 28, 2026 | Confidence |
|---|---|---|---|
| Washington | Virtix Health | Denied approximately 53% of 6,096 decisions through March. Placed on a corrective action plan in June 2026 for missed turnaround requirements. | HIGH |
| Arizona | Zyter | Approximately 14,300 decisions through March, with a denial rate just under 20%. | HIGH |
| Ohio | Innovaccer | Notified CMS before launch that the system was not ready; approved all requests automatically at first; features incomplete in April. Provider survey reports long waits and cancelled procedures. Ohio has 32 counties in the Appalachian Region (Appalachian Regional Commission). | HIGH |
| Texas | Cohere Health | No vendor-specific performance figures appear in public summaries of the release. | MED |
| New Jersey | Genzeon | No vendor-specific performance figures appear in public summaries of the release. | MED |
| Oklahoma | Humata Health | No vendor-specific performance figures appear in public summaries of the release. | MED |
Effective April 6, 2026, CMS delayed two services in all six states: deep brain stimulation for essential tremor and Parkinson’s disease, and percutaneous image-guided lumbar decompression.
The difference between Washington and Arizona deserves attention. Both operate under the same model, service list, and CMS rules, yet the reported denial rates are approximately 53 and 19 percent. The difference could reflect the populations served, the review systems, or both. CMS has not published data that would distinguish among these explanations, and until it does, questions about the reliability of these tools will remain open in regulatory, legislative, and litigation settings.
Legal framework
Discussion of AI-assisted denials frequently cites three federal cases, all currently active: Estate of Lokken v. UnitedHealth Group, Barrows v. Humana, and Kisting-Leung v. Cigna. Each rests on a contractual relationship. Lokken and Barrows involve Medicare Advantage plans and the terms of their plan documents; Kisting-Leung involves an employer-sponsored plan governed by the Employee Retirement Income Security Act (ERISA).
WISeR does not share that structure. It operates within Original Medicare under Section 1115A of the Social Security Act, with no private plan contract and no private insurer. The statute also limits administrative and judicial review of model design, including the selection of models, the selection of participants, and a model’s scope and duration (42 U.S.C. § 1315a(d)(2)). Opposition to WISeR has accordingly proceeded through FOIA, the Congressional Review Act, and the appropriations process rather than through litigation against the model. In May 2026, GAO concluded that the WISeR notice was a rule subject to the Congressional Review Act. On July 16, 2026, the Senate rejected a motion to proceed to a resolution of disapproval, 46 to 50.
Individual coverage determinations remain reviewable through the Medicare appeals process, from redetermination through an Administrative Law Judge (ALJ) hearing and, ultimately, federal district court. For that reason, the appeal record is the principal source of evidence about decision quality under the model. Overturned denials document errors, and denials that are not appealed leave errors undocumented.
This section describes the legal structure of the model for general education. It is not legal advice. Counsel should confirm the current status of any case before relying on it.
Implications beyond the pilot states
Rural Appalachia. North Carolina is not a pilot state. Ohio’s 32 Appalachian counties are, and they are served by the vendor whose launch readiness is most thoroughly documented as deficient. Automated prior authorization is therefore already operating in rural Appalachian communities in a neighboring region.
Electronic prior authorization. Under a separate rule, CMS-0057-F, many payers must support standardized electronic prior authorization beginning in January 2027. Standardized interfaces lower the cost of deploying automated review across payers, independent of WISeR’s future.
Complex rehab technology. Two WISeR categories, implanted nerve stimulators and incontinence control devices, are device categories. Medicare separately requires prior authorization for certain durable medical equipment, including some power wheelchairs, reviewed by its fixed-fee contractors. WISeR introduces a different arrangement, in which medical-necessity review is performed by a vendor compensated on reduced spending. The record does not propose applying that arrangement to complex rehab technology (CRT). If it were applied, I would expect the greatest risk of harm in CRT, for three reasons: - a custom power wheelchair is configured for one person, so an extended delay leaves the person without mobility and the supplier holding equipment it cannot deliver; - medical necessity rests on functional, narrative clinical documentation, the type of record least suited to automated criteria matching; - people who rely on this equipment are among the least likely to have the capacity to appeal.
Recommendations
For providers and suppliers, in any state: 1. Establish a baseline: days from order to authorization decision, first-pass approval rate, and the number of denials not appealed. 2. Prepare medical-necessity documentation to the standard WISeR applies, addressing each coverage criterion on first submission. 3. Appeal denials you believe are incorrect, and track outcomes. An overturn rate is evidence of decision quality. 4. Review your own AI vendor contracts for the gaps these records show: testing before go-live, turnaround commitments with defined consequences, measurable human review, and reported error rates.
For counsel: 1. Focus on individual determinations and the appeal record, which remain reviewable, rather than on model design, which largely is not. 2. Monitor the ongoing FOIA production. CMS continues to produce records, and later productions may support or qualify these findings. 3. Confirm each case citation against the docket before use.
Developments that would change this analysis
- CMS publishes vendor-level accuracy, appeal, and overturn data. That data could support the tools’ accuracy even if the response-time findings stand.
- Later FOIA productions document pre-launch testing not included in this release.
- Full-year fiscal year 2027 appropriations include the WISeR funding prohibition adopted by the House Appropriations Committee. The continuing appropriations law enacted September 2, 2026 (P.L. 119-103) does not appear to include one.
- A court interprets the review limits in Section 1115A narrowly in a procedural challenge.
I will update this article, with dated notes, as the record develops.
Sources
- CMS, “WISeR (Wasteful and Inappropriate Service Reduction) Model,” cms.gov (page modified Sept. 10, 2026); WISeR Provider and Supplier Operational Guide; WISeR Payment Methodology.
- Electronic Frontier Foundation, “New Records Reveal Problems with Medicare’s AI Prior Authorization Experiment,” Sept. 8, 2026 (updated Sept. 15, 2026), and the released WISeR FOIA Response — Combined Records, pp. 22, 99, 200, 216–217, 234, 410; Feedback Survey Responses (March 2026).
- 42 U.S.C. § 1315a (Social Security Act § 1115A), including § 1315a(d)(2); 42 U.S.C. § 1395ff and 42 C.F.R. Part 405, Subpart I (Medicare appeals).
- Government Accountability Office, decision B-337994 (May 12, 2026).
- U.S. Senate, Record Vote No. 199 (July 16, 2026), motion to proceed to S.J.Res.198, rejected 46–50; S.J.Res.192 bill status (congress.gov).
- Federal Register notice delaying two WISeR services (effective April 6, 2026).
- CMS-0057-F, Interoperability and Prior Authorization Final Rule (2024).
- Appalachian Regional Commission, Ohio state page (32 of 88 counties).
- Georgetown Health Care Litigation Tracker entries for Lokken (No. 0:23-cv-03514, D. Minn.), Barrows (No. 3:23-cv-00654, W.D. Ky.), and Kisting-Leung (No. 2:23-cv-01477, E.D. Cal.), updated Sept. 2026.
- Medicare Rights Center, “New Records Show Medicare WISeR AI Prior Authorization Model Causing Inappropriate Denials of Care,” Sept. 24, 2026; STAT, Sept. 15, 2026; Quartz, Sept. 16, 2026.
- P.L. 119-103, Continuing Appropriations and Extensions Act, 2027 (approved Sept. 2, 2026).
5Q Health provides healthcare AI governance analysis, not legal advice. This article was prepared with AI-assisted research and drafting; all findings were verified and approved by Tabatha James. Corrections are logged with dates in the site change log.