# WISeR isn't one AI tool 2026-10-26 Medicare's AI prior-auth pilot is six vendors in six states, and what the public records do and don't tell us about it. WISeR is often described as one AI system that Medicare switched on. It isn't, it's six different companies in six different states, and once you look at who runs what and how they get paid, the headlines start to look a little thin. WISeR is a [CMS Innovation Center pilot](https://www.cms.gov/priorities/innovation/innovation-models/wiser) that adds prior authorization to a short list of services in Original Medicare. That's regular Medicare, not Medicare Advantage. It runs in six states, from January 1, 2026 through the end of 2031, and CMS says it uses technology like AI and machine learning along with human clinical review. ## Six companies, not one tool Each state has its own vendor. Cohere runs Texas, Genzeon New Jersey, Humata Oklahoma, Innovaccer Ohio, Virtix Washington, and Zyter Arizona. So when someone says "the WISeR AI denied X," my first question is which one. I couldn't find a public description of how any vendor's WISeR system works. All I can say about the technology is what CMS says about it. ## Who has to review a "no" CMS's [payment methodology](https://www.cms.gov/priorities/innovation/files/wiser-payment-methodology.pdf) says every non-affirmation, which is their word for a request that doesn't get approved, has to get a medical review by a licensed clinician. That's on paper, though, and how much that review changes the outcome isn't public. Also, a "denial" here is a decision on a request, not on a claim, and providers can resubmit as many times as they want. ## How the vendors get paid If you only read one section of the CMS paperwork, I'd make it this one. Per CMS's [participant guide](https://www.cms.gov/priorities/innovation/files/wiser-participant-guide.pdf), vendors earn 25% of an average regional spending figure, after a discount, for each request they decline that isn't later overturned. A quality score can adjust that payment down by 0 to 10%. If a denial gets overturned on resubmission or appeal, they don't get paid for it. Before the final design, CMS's own actuaries looked at an early draft. In a June 2025 memo, released through [EFF's records lawsuit](https://www.eff.org/files/2026/09/07/combined_records_-_2nd_interim_release.pdf), they wrote that "model participants will have an incentive to deny as many claims as possible." That memo is about the draft design, not the final one, and it says nothing about what any vendor actually did. But it's CMS's own people saying it, so I'm not going to skip it. ## What the first numbers show Records released in that lawsuit show Washington's vendor denied 53% of the 6,096 requests it had reviewed by late March. A different vendor in Arizona denied under 20% of about 14,300. [Ars Technica](https://arstechnica.com/health/2026/09/trump-admin-using-ai-to-deny-medical-care-for-seniors-in-disastrous-experiment/) reported Washington's numbers. Arizona's figure comes from the [Seattle Times](https://www.seattletimes.com/business/medicare-ai-program-hit-wa-with-staggering-denial-rate-records-show/) alone, which doesn't name the vendor. I'd be careful with these. They're the vendors' own weekly reports to CMS, not audited results, and they cover only the first weeks of the program. The vendors may also count things differently, since nobody says whether resubmissions are counted, and Arizona's exact rate was never published. It's the same pilot with the same rules from CMS, and the early numbers are very different, which may say something about how each operator and its process works. It doesn't tell us how much of the gap is the AI itself, and I couldn't find anything public that says. I'd call it a hypothesis, not a finding. ## The Texas numbers You'll see Texas numbers too. Early reports, tracing back to a March Washington Post story, said officials told them Texas's vendor initially approved about 62% of requests, rising to about 84% once a human reviewed them. I'm going off secondhand coverage, [EFF's court complaint](https://www.eff.org/files/2026/03/25/complaint_-_eff_v_cms.pdf) and [Healthcare Dive](https://www.healthcaredive.com/news/electronic-frontier-foundation-sues-cms-medicare-ai-prior-authorization-wiser/816087/), because the original is paywalled. The sources don't say what "initially" or "human review" meant, so I wouldn't read much into the gap. ## Waits Records document waits of several weeks and at least one request pending 83 days. Washington hospitals told [Fierce Healthcare](https://www.fiercehealthcare.com/regulatory/ai-powered-prior-authorizations-medicare-have-greatly-delayed-care-wash-hospitals-say) it was taking four to eight weeks where about two was normal. One 83-day record isn't a rate, and some of the time is routing through Medicare's regional contractors, not only the vendors. ## Congress and CMS On July 16 the Senate voted 46 to 50, along party lines, against [taking up a resolution](https://www.senate.gov/legislative/LIS/roll_call_votes/vote1192/vote_119_2_00199.htm) to end WISeR. That was a procedural vote, not a verdict on the program. In October, CMS Deputy Administrator Kimberly Brandt said, per [Nextgov](https://www.nextgov.com/artificial-intelligence/2026/10/cms-hopes-quadruple-provider-suspensions-over-next-six-months-official-says/416640/): "There are a lot of things about WISeR that are not working." She didn't say which things. ## How big is it The pilot runs in six states with about 6.4 million Original Medicare beneficiaries in 2024, but it applies only to a short list of services. [KFF](https://www.kff.org/medicare/examining-the-potential-impact-of-medicares-new-wiser-model/) puts the number of people there who used one of those services in 2024 at about 207,500. ## The other prior-auth rule There's also a separate federal prior-authorization rule, [CMS-0057-F](https://www.cms.gov/newsroom/fact-sheets/cms-interoperability-prior-authorization-final-rule-cms-0057-f). That one covers Medicare Advantage, Medicaid, CHIP and federal-exchange plans, generally from 2027. Original Medicare isn't on its list, so WISeR sits outside it. ## Where that leaves me I don't have a clean takeaway yet. The public record shows a pilot with six operators, a draft payment design that CMS's own actuaries said gives an incentive to deny, and early numbers that can't be compared cleanly. What I'd want next is audited data and a plain explanation of what each vendor's system does before anyone says what the AI is or isn't doing. ## Sources - CMS WISeR model page: https://www.cms.gov/priorities/innovation/innovation-models/wiser - CMS WISeR FAQ: https://www.cms.gov/priorities/innovation/files/document/wiser-model-frequently-asked-questions - CMS payment methodology: https://www.cms.gov/priorities/innovation/files/wiser-payment-methodology.pdf - CMS participant guide: https://www.cms.gov/priorities/innovation/files/wiser-participant-guide.pdf - EFF records release (actuary memo): https://www.eff.org/files/2026/09/07/combined_records_-_2nd_interim_release.pdf - EFF complaint v. CMS (Texas figures, para. 15): https://www.eff.org/files/2026/03/25/complaint_-_eff_v_cms.pdf - Healthcare Dive on the EFF suit: https://www.healthcaredive.com/news/electronic-frontier-foundation-sues-cms-medicare-ai-prior-authorization-wiser/816087/ - Ars Technica: https://arstechnica.com/health/2026/09/trump-admin-using-ai-to-deny-medical-care-for-seniors-in-disastrous-experiment/ - Seattle Times: https://www.seattletimes.com/business/medicare-ai-program-hit-wa-with-staggering-denial-rate-records-show/ - Fierce Healthcare: https://www.fiercehealthcare.com/regulatory/ai-powered-prior-authorizations-medicare-have-greatly-delayed-care-wash-hospitals-say - Senate roll call 199: https://www.senate.gov/legislative/LIS/roll_call_votes/vote1192/vote_119_2_00199.htm - Nextgov: https://www.nextgov.com/artificial-intelligence/2026/10/cms-hopes-quadruple-provider-suspensions-over-next-six-months-official-says/416640/ - KFF on WISeR: https://www.kff.org/medicare/examining-the-potential-impact-of-medicares-new-wiser-model/ - CMS-0057-F fact sheet: https://www.cms.gov/newsroom/fact-sheets/cms-interoperability-prior-authorization-final-rule-cms-0057-f