On March 9, 2026, a federal judge ordered UnitedHealth Group to hand over the internal documents behind its AI denial algorithm — the same tool used to refuse post-acute care to Medicare Advantage patients at a rate 9x higher than before the algorithm launched.
This wasn’t a regulatory warning. It was a federal court order. And most physicians have no idea it happened.
If you have Medicare Advantage patients, AI is already rationing their care. If you’re in one of six states, it arrived in traditional Medicare on January 1. And it is designed so that the vast majority of patients wrongly denied will never bother appealing.
The AMA’s 2026 survey is unambiguous: 61% of physicians say AI is increasing prior authorization denials, and the Senate investigation found Humana’s AI-driven post-acute denial rate ran 16 times its own baseline. The good news: 75% of AI-generated Medicare Advantage denials that are appealed get overturned. The bad news: fewer than 0.2% of policyholders ever appeal — (STAT News, March 23, 2026).
That asymmetry is not a bug. It is the business model.
Here’s who is running what AI against you, what changed in January 2026, and exactly what to do when an algorithm blocks your patient’s care.
Three Events That Changed Prior Authorization in 2026
Three things happened in early 2026 that should have been front-page clinical news. They largely weren’t.
March 9, 2026: Federal Judge John R. Tunheim (U.S. District Court, District of Minnesota) ordered UnitedHealth Group to produce internal records on nH Predict — including its design specifications, its connections to cost-saving strategy, and government investigation documents. Class action plaintiffs allege a 90% wrongful denial rate. UnitedHealth’s position is that nH Predict is a “guide,” not a decision-maker. A federal judge found that argument worth examining more closely.
January 1, 2026: CMS launched the WISeR (Wasteful and Inappropriate Service Reduction) pilot in six states — New Jersey, Ohio, Oklahoma, Texas, Arizona, and Washington. For the first time, AI-screened prior authorization arrived in traditional Medicare, across 17 service categories including spinal procedures, neurostimulation, and knee arthroscopy. If you practice in those states and bill any of those codes, you are in it now, whether you know it or not.
March 23, 2026: Writing in STAT News, Dr. Oni Blackstock surfaced the figure that reframes the entire prior authorization fight: 75% of AI-generated Medicare Advantage denials that are appealed are overturned. The system generates massive volumes of incorrect denials. It just counts on the fact that you and your patient won’t have time to challenge them.
Who’s Running AI Against You — and How Their Tools Actually Work
Two payer-side systems account for a disproportionate share of AI prior authorization denials doctors are fighting. Physicians should know how they work.
nH Predict is NaviHealth’s algorithm, now owned by Optum and UnitedHealth Group. It estimates how many post-acute care days a patient “should” need — skilled nursing facility stays, home health visits, inpatient rehabilitation — based on a proprietary model. UnitedHealth calls it a guide for clinical reviewers. The numbers tell a different story: skilled nursing facility denial rates rose from 1.4% in 2019 to 12.6% in 2022, the first full year NaviHealth managed claims (CBS News / class action coverage). That is not a guide being used cautiously.
eviCore (owned by Evernorth/Cigna) handles prior authorization for approximately 100 million consumers — roughly 1 in 3 insured Americans (ProPublica). Its internal AI system is reportedly called “the dial.” According to ProPublica’s investigation, it can be tuned to route more claims to in-house reviewers — where denial rates are higher — and salespeople have pitched prospective insurer clients on “15% increases in denials” as a selling point.
The Senate Permanent Subcommittee on Investigations (October 2024) examined all three largest Medicare Advantage insurers — UnitedHealth, Humana, and CVS/Aetna — and found post-acute AI denial rates running 3 to 16 times above their own baselines (Healthcare Dive).
And UnitedHealth CEO Steven Hemsley has publicly pledged to use AI to cut $1 billion in costs in 2026 (STAT News, February 2026). Your patients’ denied claims are a line item in that math.
Here’s what we think: These tools have never undergone anything resembling FDA-level validation. They make de facto medical necessity decisions on millions of patients with no clinical accountability. As interventional radiologist Dr. Zeke Silva, MD put it in Medscape: “We accept that prior authorization has some role. We accept that algorithms have some role in patient care. But if you’re denying care, it should be by someone trained to practice medicine.” The same rigor demanded for a diagnostic AI — peer-reviewed validation, disclosed error rates, outcome data — should apply to any AI that denies care. None of it is currently being demanded, and payers are not volunteering it.
What the AMA’s 2026 Data Actually Shows (Beyond the Headline)
The 61% headline is damning. The full AMA 2026 survey dataset is worse.
| Metric | Finding |
|---|---|
| Physicians who say AI is increasing PA denials | 61% |
| Physicians who say PA negatively impacts clinical outcomes | 94% |
| Physicians who say PA increases burnout | 89% |
| Physicians who’ve had a patient suffer a serious adverse event due to PA | 29% |
| Average hours per week spent on prior authorization (physician + staff) | 13 hours |
That 29% figure deserves more attention than it gets. Nearly one in three physicians surveyed report at least one patient who was hospitalized, suffered permanent damage, or died because a prior authorization was delayed or denied. These are not inconveniences. They are outcomes (AMA 2026 Survey).
On the 16x statistic: it is frequently misquoted. The precise finding from the Senate Subcommittee is that Humana’s AI-driven post-acute care denial rate ran 16x its own overall prior authorization baseline — not an industry-wide comparison, and not compared to a peer institution. Humana’s AI was running 16 times hotter than Humana’s own standard. That is the number.
Dr. Leslie Lenart, MD, a primary care physician and computing researcher at the Medical University of South Carolina, put the technical problem plainly in Medscape: “These kinds of if/then algorithms are too fragile for the complex reasoning it takes to evaluate claims.”
On the data war: Two weeks after the AMA’s 61% finding, Cohere Health — a payer-contracted PA platform — released a self-commissioned survey showing 99% of clinicians confident in AI-driven prior authorization. AJMC flagged the conflict explicitly (AJMC). When the AMA surveys 1,000+ physicians and gets 61% concern, and a payer-contracted vendor surveys 200 clinicians and gets 99% confidence, the difference is the research design — not reality. Know who commissioned what before you cite it.
CMS WISeR: What It Is, What It Isn’t, and What Physicians in Six States Must Do Now
WISeR stands for Wasteful and Inappropriate Service Reduction. The name tells you whose perspective it was designed from.
The program launched January 1, 2026 and runs through 2031 across New Jersey, Ohio, Oklahoma, Texas, Arizona, and Washington. It covers traditional Medicare only — Medicare Advantage and Medicaid are not in scope. It requires prior authorization for 17 service categories, including:
- Neurostimulation devices
- Spinal procedures
- Knee arthroscopy for osteoarthritis
- Skin and tissue substitutes
- Incontinence control devices
- Certain nerve stimulation services
Participation is technically voluntary. But non-submitters face automatic prepayment review — meaning CMS can hold your reimbursement while reviewing your claims. For any practice that bills these codes with any regularity, voluntary is a legal fiction.
AI is used for initial screening. CMS requires that all non-payment recommendations be reviewed by an appropriately licensed clinician before denial. That is a meaningful safeguard on paper. In practice, the AMA’s data on Medicare Advantage exists in a world where human review was also nominally required — and still produced denial rates 3–16x baseline. The safeguard is only as strong as the time and expertise of the reviewer the vendor assigns.
If you’re in a WISeR state, know your assigned vendor:
| State | WISeR Vendor |
|---|---|
| Texas | Cohere Health |
| New Jersey | Genzeon |
| Oklahoma | Humata Health |
| Ohio | Innovaccer |
| Washington | Virtix Health |
| Arizona | Zyter |
Submit through your WISeR vendor portal or your Medicare Administrative Contractor (MAC). Full service category list and submission details: cms.gov/priorities/innovation/innovation-models/wiser.
The Tools That Claim to Be on Your Side (and Their Limitations)
The speed asymmetry is stark: 84% of insurers use AI or machine learning for prior authorization decisions, while most physician practices — particularly solo and small group — still submit manually (MedCity News, February 2026).
Provider-facing tools exist. Their quality and independence vary significantly.
Humata Health serves 42,000 physicians across 226 facilities and claims a 90% touchless approval rate. It integrates with Microsoft Dragon Copilot for in-workflow PA automation. Humata is also the WISeR vendor for Oklahoma — which means it serves both sides of the transaction in that state.
Innovaccer Flow Auth auto-drafts appeal letters using payer-specific policies and historical outcome data. It is the WISeR vendor for Ohio.
Silna handles benefit verification and PA automation across 1,000+ payers, with a fully provider-facing model. Its CEO Jeffrey Morelli has made the structural argument clearly: “Refusing to adopt it guarantees the status quo: a system where insurers operate at machine speed, providers operate at human speed.”
Cohere Health offers a provider portal but is primarily a payer-contracted platform — health plans are the paying clients. Its reported 83% immediate approval rate was reported by Cohere, for plans that contracted Cohere. Independent outcome data does not exist.
The determinative question to ask any PA AI vendor: Who pays you? If the answer is a health plan, that plan’s interests come first. This is not cynicism — it is contract logic.
For physicians interested in reducing documentation burden without ceding clinical judgment, AI medical scribes for doctors represent a genuinely different category: tools that remove administrative work without touching care decisions. That distinction matters. The 13 weekly hours physicians spend on prior authorization is the problem worth solving. AI clinical decision support tools that actually work for clinicians are another space worth examining separately — the evidence base is better and the conflicts of interest are cleaner.
What to Do Right Now When AI Denies Your Patient’s Care
When P. Dileep Kumar, MD described a medication renewal denial for a patient stable on the treatment for years, he wrote: “An algorithm, silent, opaque, and final. Denied: automated appropriateness determination — no reviewer, no rationale, no appeal path.”
That description has become accurate at scale. Here is how to fight it.
1. Get the denial in writing immediately. Do not accept a verbal denial. Request the specific clinical criteria used and whether a named physician reviewer was involved. If they can’t name one, that is the beginning of your appeal.
2. Ask directly whether AI was used. New CMS transparency requirements effective 2026 require disclosure of AI use in utilization review. Document the answer — this matters for both your appeal and any complaint you file later.
3. Request a peer-to-peer with a physician in your specialty. Not a general medical reviewer. Document their name, their specialty, and the time they spent on your case. A five-minute review of a complex oncology denial is itself evidence of inadequate review.
4. Draft the appeal with specific language. Reference your specialty’s clinical guidelines for medical necessity. Include patient history of failed alternatives. Document the risk of deterioration if care is delayed. Generic appeals lose. Specific, cited appeals get overturned.
5. Cite the 75% overturn rate in your appeal letter. It signals that you understand the data and will not drop the matter. AI denials that are challenged are overturned at very high rates — the insurer knows this too.
6. Know your state’s protections. Texas and Arizona have legislation requiring human physician review for clinical determinations — check current status. If your state has an external appeal mechanism, use it.
7. File a CMS complaint for any Medicare Advantage denial where no named physician reviewer can be identified. The Office of Inspector General is actively investigating AI-assisted denial practices in MA. Your complaint becomes part of that record. The complaint pathway is at CMS.gov.
A note on patient follow-up: If you’re continuing this fight via telehealth, direct your patient to prepare for the telehealth appointment with the denial letter, appeal notes, and timeline of events already in hand. It prevents the entire visit from being spent reconstructing paperwork.
Frequently Asked Questions
Is AI actually making prior authorization worse for doctors, or just shifting the burden?
Both — and the AMA data confirms it. 61% of physicians report AI is increasing denials, not just automating the same denial rate faster. The 13 weekly hours spent on PA have not decreased. For the average practice, payer AI has made fighting denials harder: the denial comes back faster, the criteria are less transparent, and the appeals pathway is more opaque. If 75% of AI denials that are appealed get overturned, the system is generating enormous volumes of incorrect denials and counting on physicians and patients not having time to fight them.
Which AI prior authorization tools are designed to help physicians vs. payers?
Payer-side tools (used to review and deny): nH Predict (UnitedHealth/NaviHealth) and eviCore (Cigna/Evernorth). Provider-side tools (with caveats): Humata Health, Innovaccer Flow Auth, and Silna. Cohere Health occupies a middle position — primarily payer-contracted but with a provider portal. The question to ask of every tool is the same: who is the paying client? The answer tells you whose interests the algorithm is optimizing.
What does the AMA’s 2026 data actually say about AI and prior authorization?
61% of physicians say AI is increasing PA denials. 94% say PA negatively impacts clinical outcomes. 89% say PA increases burnout. 29% have had a patient suffer a serious adverse event — hospitalization, permanent damage, or death — due to a PA delay or denial. The widely cited 16x figure is Humana-specific: it represents Humana’s AI-driven post-acute care denial rate compared to Humana’s own overall PA baseline, from the U.S. Senate Permanent Subcommittee on Investigations report (October 2024) — not an industry average.
How do you appeal a prior authorization denial that was generated by an AI system?
Get the denial in writing and ask whether AI was used — CMS now requires disclosure. Request a peer-to-peer with a physician in your specialty, not a general reviewer. Appeal using medical necessity language that cites your specialty’s clinical guidelines — generic appeals fail. Reference the 75% MA appeal overturn rate in your letter. If no named physician reviewer can be identified for a Medicare Advantage denial, file a CMS complaint. Document every step.
What is the CMS WISeR program and does it help or hurt clinicians?
WISeR launched January 1, 2026 in six states (NJ, OH, OK, TX, AZ, WA), applying AI-screened prior authorization to traditional Medicare for the first time across 17 service categories. CMS requires human clinician review before any denial. Critics — including the AMA — note that “human review” in prior authorization contexts can mean a physician outside your specialty reviewing an AI summary under time pressure, which is structurally similar to what produced 3–16x denial rate increases in Medicare Advantage. If you’re in a WISeR state and bill any of the 17 flagged categories, submit through your assigned WISeR vendor or MAC to avoid automatic prepayment review.
The Bottom Line
AI prior authorization is not neutral. It is currently deployed far more aggressively against your patients than it is being used to help you — and the evidence for this is now in federal court documents, a Senate subcommittee report, and the clinical records of the 29% of physicians who watched a patient deteriorate while a prior authorization was pending.
If you have Medicare Advantage patients or practice in a WISeR state: document every AI-generated denial, demand to know whether a named physician reviewed it, and appeal. Because 75% of the time, when someone bothers, the algorithm turns out to have been wrong.
The biggest AI promise in healthcare was always reducing paperwork. For the physician spending 45 minutes fighting a three-second algorithmic denial, it has done exactly the opposite.
Sources: AMA 2026 Survey · Senate PSI Report, October 2024 · STAT News, March 23, 2026 · STAT News, February 2026 · ProPublica on eviCore · CBS News / nH Predict class action · AJMC on Cohere Health data conflict · MedCity News, February 23, 2026 · CMS WISeR model page · Medscape on prior auth AI, April 2025 · KevinMD, December 2025 · KevinMD, April 2025
