ARCLIGHT

Search Arclight

Find services, research, and practical tools.

Popular topics

    Arclight Insight / November 16, 2026

    Medicare's Clinical Superintelligence Roadmap Already Has a Test Case

    CMS wants clinical superintelligence (SI) to earn greater autonomy. The WISeR rollout shows what Medicare should require before it lets these systems shape access to care at scale.

    Explore the interactive roadmap ↓

    By Lance McNeill | Arclight Action | November 2026

    WISeR, the Wasteful and Inappropriate Service Reduction Model, is a CMS Innovation Center model that began January 1, 2026, in Arizona, New Jersey, Ohio, Oklahoma, Texas and Washington. In Traditional Medicare, six private technology companies use AI and machine learning, along with clinician review, to screen selected services CMS considers vulnerable to waste, fraud and abuse. Providers can request prior authorization before treatment; if they do not, the claim goes to prepayment review. The companies are paid a percentage of the spending CMS counts as averted by their reviews.

    WISeR's first performance period closed on March 31. Under the timeliness measure CMS wrote for that period, a prior authorization request or prepayment review that "does not have a determination within the performance period" is excluded from the calculation. So are dismissed requests and ineligible prepayment reviews. The measure counts decisions issued, not people waiting.

    The following week, after that period had closed, CGS, Ohio's Medicare contractor, filed its status report for the week of April 6. That operating snapshot counted 530 prior authorization submissions on hold pending a determination from Innovaccer, Ohio's WISeR participant, and 1,874 claims suspended for prepayment review. The oldest suspended claim was 77 days old. Those figures are not a count of unique patients. They show what a measure built on completed decisions can miss.

    CMS has published its Clinical Super Intelligence (SI) Roadmap, a federal strategy for advanced AI in health care. It holds these tools to a "better than today" standard and says "expanded autonomy must be earned through evidence, staged deployment, and continuous real-world monitoring."

    How will CMS know whether AI makes care faster if its measures can leave out the cases that are still waiting?

    A program CMS already runs

    The roadmap does not mention WISeR or prior authorization. Its scope runs from patient and clinician tools to potential AI-led primary care and AI-led insurance options. CMS itself has said WISeR "can provide a roadmap for incorporating more private sector innovations into CMS operations."

    My September case study documented what broke in that rollout. Who Was Watching WISeR? asked who was responsible for seeing it. This piece asks the next question: If CMS applied its own new standard to WISeR, what would Medicare have to prove before expanding it?

    Open the interactive roadmap in a new tab

    Count everyone who is waiting

    The roadmap promises "real-world, real-time evidence systems so performance can be tracked after tools are deployed." For WISeR, that should mean publishing a count of every request and claim still waiting next to the timeliness score. The count should show how long each has waited and, for pending prior authorizations, the planned date of service.

    CMS's August specification said timeliness would be scored on a pay-for-reporting basis "through at least PP3," the performance period that ended September 30. Publishing the size and age of the queue does not have to wait for performance scoring.

    Make staged deployment a public decision

    The roadmap says it "does not authorize autonomy that has not cleared the 'better than today' bar for its risk tier." Ohio shows why that bar has to be visible.

    On December 22, CGS reported its first full loop of prior authorization testing with Innovaccer. End-to-end prepayment testing was unfinished, and a text-recognition defect was sending incorrect beneficiary identifiers. CGS wrote that part of the participant's solution "will not be available at go live," with next steps "pending decision from CMS."

    Prior authorization intake opened January 5. On April 6, CGS reported that Innovaccer still had not completed full testing of prepayment review or established a process for issuing non-affirming prior authorization decisions.

    Under the roadmap's logic, the December decision should have been a public one. Before any higher-stakes launch or expansion, CMS should publish what had to work, what failed testing, which risks it accepted, and who signed off.

    Give human escalation a job

    The roadmap expects AI-led care organizations to build "partnerships with clinicians and facilities for appropriate escalation to human clinicians."

    In Texas, I worked on a skin-substitute claim for a pressure ulcer outside the lower extremity that was suspended for WISeR prepayment review, even though diagnosis and wound site determine whether that service is within the model's scope.

    In an October 8 reply to me, CMS's WISeR help desk confirmed the general rule: such a claim falls outside the model, and scope is checked before any medical-necessity review. For the individual claim, the help desk directed me to Novitas, the Medicare contractor for Texas: "Case level confirmation must be obtained through the billing provider's MAC." It added, "The WISeR Help Desk is unable to validate these details."

    That reply settled the rule. It did not settle the claim. Someone needs the authority to take a misrouted claim out of review, get it moving again, and tell the practice when to expect a usable answer.

    In WISeR, the speed sits on the machine side. Cohere, the Texas participant, says its AI can approve a request "within 10 or 15 seconds"; denials go to its clinicians. The delay sits on the human side. A non-affirmed prior authorization can be resubmitted with peer-to-peer review, but a denied claim enters Medicare's standard appeals process. A contractor's redetermination and an independent reconsideration of the written record each generally take up to 60 days. A hearing, usually by telephone, comes at the third level, before an administrative law judge with 90 days to decide; the fiscal 2026 average has been 69 days. With filing windows, months can pass before a provider speaks with anyone who can decide the case.

    That was my experience with Medicare appeals: months before I reached a human reviewer I could actually talk to, someone who would have an honest conversation, let me point out the problem, and listen. Not every WISeR denial takes that long. But when the AI answers in seconds and the human answers in months, the system is fast in only one direction.

    The same test applies when a patient needs an AI-led service to connect them with a clinician: a named owner, a deadline, a fallback, and a measured time from first request to resolution.

    Measure who went without care

    The roadmap is right to point payment toward results. It cites the ACCESS model, which "rewards clinical results rather than activities." Under ACCESS, full payment depends on outcomes "such as improvement or control of blood pressure for hypertension." Beneficiaries lost to follow-up stay in the denominator of its outcome calculation. ACCESS, in other words, counts the people who drop out.

    WISeR rewards something else: savings CMS attributes to qualifying non-affirmations and prepayment denials, withheld or recouped when a denial is overturned on appeal.

    Hannah Neprash and Michael Barnett published the first independent study of WISeR's effects, a JAMA research letter, on October 26. Their estimates are comparative. Relative to trends outside the pilot states, traditional Medicare encounters for epidural steroid injections fell 15.0% in the first six months, and the other targeted services, pooled, fell 19.5%. The study could not determine whether the forgone care was needed, or observe how AI was used.

    The roadmap itself warns that SI systems "may also be shaped by financial incentives." A model that rewards lower spending should track delayed and abandoned care as carefully as ACCESS tracks patients lost to follow-up. CMS has already asked its WISeR evaluator to study "delays in medically necessary care, abandonment of care, avoidable hospital admissions, and adverse events." Those findings should be public before the model grows.

    The test CMS can answer

    Before expanding WISeR, CMS should publish a readiness and performance finding for each participant and Medicare contractor connection: six participants working with three Medicare Administrative Contractors. Each finding should show:

    1. Which workflows passed end-to-end testing, and which did not.
    2. How many requests and claims are still waiting, and for how long.
    3. How many cases were dismissed or found ineligible, and how long correction took.
    4. What CMS did when a participant or contractor missed its standards.
    5. Who can pause a failing pathway, what triggers a pause, and where cases go in the meantime.

    CMS already holds the pause power. Each WISeR participation agreement lets CMS "suspend or terminate" a participant's prior authorization process on new requests, including when a participant has taken action that "threatens the health or safety of a Beneficiary." That section assigns the power to "CMS" without naming who exercises it or saying where pending cases go. Arclight recommends that CMS name the Innovation Center official accountable for each pathway, publish the backlog and error thresholds that trigger a pause, and return affected requests and suspended claims to the MAC for ordinary Medicare processing until the pathway passes retesting.

    That is the roadmap's own test: transparent evidence, staged deployment, human oversight and continuous monitoring. A patient whose request is still on hold belongs in Medicare's account of how well the system works.

    Sources

    Page numbers are one-based PDF positions unless noted.

    1. CMS, "WISeR Performance Measure Technical Specifications for Performance Period (PP) 1," August 13, 2026, pp. 2–3 (denominator exclusions; pay-for-reporting through at least PP3). https://www.cms.gov/priorities/innovation/files/wiser-perf-msr-tech-spec.pdf
    2. EFF, "Combined Records, 2nd Interim Release," pp. 200–201 (CGS status report, week of April 6, 2026); pp. 220–221 (CGS status report, week of December 22, 2025). https://www.eff.org/files/2026/09/07/combined_records_-_2nd_interim_release.pdf
    3. "A Clinical Super Intelligence Roadmap: Better Answers, Faster Care, and Lower Costs," posted by CMS, November 2026, pp. 1–2, 4, 6–8. https://www.cms.gov/files/document/clinical-si-roadmap.pdf
    4. CMS, WISeR Model page (participants, MAC jurisdictions, review routes, participant payment). https://www.cms.gov/priorities/innovation/innovation-models/wiser
    5. CMS, WISeR Participant Guide, Table 3 (performance period dates). https://www.cms.gov/priorities/innovation/files/wiser-participant-guide.pdf
    6. CMS, WISeR Model Frequently Asked Questions (January 5, 2026 intake for services on or after January 15). https://www.cms.gov/priorities/innovation/files/document/wiser-model-frequently-asked-questions
    7. Arclight Action, "WISeR and the Human Cost of Testing AI in Production," September 2026. https://www.arclightaction.com/insights/wiser-ai-rollout-case-study
    8. Arclight Action, "Who Was Watching WISeR?," November 2, 2026, citing Mathematica task order 75FCMC26F0017, p. 6, in EFF, "Additional Applications and Contracts." https://www.arclightaction.com/insights/who-was-watching-wiser; https://www.eff.org/document/wiser-foia-response-additional-applications-and-contracts
    9. Arclight Action, "When a WISeR ADR Is Outside the Model's Scope." https://www.arclightaction.com/insights/wiser-adr-outside-model-scope
    10. CMS WISeR help desk response to Lance McNeill, October 8, 2026 (ticket CS6056353). Private correspondence; public rule in CMS, WISeR Provider and Supplier Operational Guide, § 4.1.4. https://www.cms.gov/priorities/innovation/files/wiser-provider-supplier-guide.pdf
    11. CMS, ACCESS Model Technical FAQ. https://www.cms.gov/priorities/innovation/access-technical-frequently-asked-questions
    12. CMS, ACCESS payment amounts and performance targets, p. 6 (outcome attainment rate; beneficiaries lost to follow-up remain in the denominator). https://www.cms.gov/priorities/innovation/files/access-payments-amts-perf-targets.pdf
    13. CMS, WISeR Model Payment Methodology. https://www.cms.gov/priorities/innovation/files/wiser-payment-methodology.pdf
    14. Neprash HT, Barnett ML, "Changes in Service Use Following the Introduction of AI-Enabled Prior Authorization in Medicare," JAMA, October 26, 2026, doi:10.1001/jama.2026.21230. https://pubmed.ncbi.nlm.nih.gov/42932180/; JAMA Network media release: https://media.jamanetwork.com/media-releases/10.1001/jama.2026.21230
    15. University of Minnesota School of Public Health, "Medicare's prior-authorization pilot test led to an immediate drop in use of services, study finds," October 26, 2026. https://www.sph.umn.edu/news/medicares-prior-authorization-pilot-test-led-to-an-immediate-drop-in-use-of-services-study-finds/
    16. DocWire News, "Medicare's AI Prior Auth Pilot Tied to 15%-20% Drop in Targeted Services." https://www.docwirenews.com/post/medicares-ai-prior-auth-pilot-tied-to-15-20-drop-in-targeted-services
    17. EFF, "Participation Agreements" production, pp. 60–61 (Article XIII, §§ 13.1–13.2; the same terms appear in each of the six agreements). https://www.eff.org/files/2026/09/07/r_combined_records_wiser_pas.pdf
    18. CMS, "Original Medicare (Fee-for-service) Appeals," and pages for the first, second and third levels of appeal (filing windows; redetermination and reconsideration generally within 60 days; ALJ hearings generally by telephone; 90-day ALJ adjudication period). https://www.cms.gov/medicare/appeals-grievances/fee-for-service; https://www.cms.gov/medicare/appeals-grievances/fee-for-service/first-level-appeal-redetermination-medicare-contractor; https://www.cms.gov/medicare/appeals-grievances/fee-for-service/second-level-appeal; https://www.cms.gov/medicare/appeals-grievances/fee-for-service/third-level-appeal
    19. HHS Office of Medicare Hearings and Appeals, "Average Processing Time by Fiscal Year" (FY2026: 69.0 days; content last reviewed August 11, 2026). https://www.hhs.gov/about/agencies/omha/about/current-workload/average-processing-time-by-fiscal-year/index.html
    20. Cohere Health, "AI in Prior Authorization: Access & Cost Benefits" (company statement about its platform generally, not WISeR-specific). https://www.coherehealth.com/news/smarter-prior-authorization-ai-care-access-cost-reduction

    AI Assurance

    Arclight provides independent assessment of high-stakes AI systems, including governance, transparency, human review, access, and accountability.

    Explore AI Assurance