Prior authorization is the part of medicine that most patients never see, but most clinicians are fighting every single day. As an anesthesiologist and pain medicine physician, I have spent more clinical hours than I care to count on authorizations for the pain procedures I do. What results is care that is often delayed by weeks for patients who are already suffering. AI is now being deployed on both sides of this fight: by payers to screen and deny requests faster, and by providers to submit them faster and fight denials more effectively. This week, we examine who is winning, what the law is starting to say about it, and what clinicians can do now.

In today's newsletter:

Prior authorization was designed as a cost-control mechanism: require clinical justification before approving expensive or potentially unnecessary care. In theory, reasonable. In practice, a system that delays care for 93% of physicians' patients, contributes to serious adverse events in nearly 30% of cases by physician report, and consumes an estimated two full business days of physician time per week. In pain medicine and anesthesiology, the procedures most commonly targeted by PA — spinal cord stimulators, implantable pumps, certain nerve blocks, opioid alternatives — are precisely the ones that represent the evidence-based alternative to medications payers claim to want patients on instead. The administrative obstruction is not a bug in the system. For certain payers, it is a revenue strategy.

AI entered this space from both directions simultaneously. On the payer side, insurers built algorithmic screening tools designed to identify requests that could be denied based on protocol mismatches, incomplete documentation, or statistical outlier status — without requiring a physician reviewer to read the chart. UnitedHealth's nH Predict algorithm, the subject of ongoing litigation, was found to have a 90% error rate on appeal, meaning nine out of ten denials that patients and physicians challenged were ultimately reversed. The algorithm was right less often than random chance on the cases that went to appeal. On the provider side, companies like Cohere Health, PrescriberPoint, and now Abridge are building tools that compile the clinical documentation, match it against current payer criteria, and submit authorizations automatically, sometimes before the visit ends.

The structural question this creates is not whether AI should be involved in prior authorization because it already is, on both sides. The question is who bears accountability when an AI denial causes a patient harm, and whether the current regulatory framework is equipped to answer it. Four states passed laws in 2025 prohibiting payers from using AI as the sole basis for a PA denial without physician review. CMS launched its WISeR program in January 2026, testing AI-assisted review in six states for specific high-cost procedures. And the White House's March 2026 AI policy framework is attempting to preempt those state laws before they spread further. The fight over who controls the prior auth process is also a fight over who controls healthcare AI accountability.

There is a growing national effort to modernize prior authorization. CMS Administrator Dr. Mehmet Oz announced a coalition of nearly 30 healthcare organizations committed to replacing fax based workflows with digital prior authorization and standardized data exchange. The emphasis is on reducing administrative burden rather than increasing denials, which is an important distinction as clinicians remain understandably skeptical.

What it means: If successful, electronic prior authorization could return meaningful clinical time back to physicians instead of staff spending hours navigating payer requirements. This is huge for anyone who still has to fax anything in. for authorization

A federal court ordered broad discovery against UnitedHealth Group in the class action lawsuit over its nH Predict AI algorithm, which was used to determine post-acute care length of stay for Medicare Advantage patients. The algorithm's error rate on appealed denials was found to be approximately 90% — meaning nearly every denial that was challenged was ultimately reversed. Class certification is scheduled for 2026. A 2023 congressional investigation found UnitedHealthcare denied approximately 32% of prior authorization requests for Medicare Advantage members, a rate significantly higher than other major insurers.

What it means: The legal and regulatory pressure on payer-side AI denial tools is real and accelerating. When patients and physicians appeal AI-generated denials, they win nine times out of ten. Appeal every denial that affects your patients — the data supports it.

Abridge and Availity announced a partnership at JPM26 to integrate real-time prior authorization directly into the ambient documentation workflow, compressing what is typically a multi-week process into a real-time transaction during the patient visit. Abridge's contextual reasoning engine captures clinical dialogue and structures it into the documentation required for authorization; Availity's payer connectivity then submits that request in real time. The integration is projected to support over 80 million clinical conversations in 2026.

What it means: This is what provider-side AI is supposed to look like: removing the administrative lag between the clinical decision and the insurance approval. If your health system uses Abridge, ask whether the Availity integration is in your deployment roadmap.

American Medical Association, 2025-2026

The AMA's ongoing national survey on prior authorization found that 29% of physicians reported PA delays or denials led to a serious adverse event for a patient, 23% reported a patient hospitalization attributable to PA obstruction, and 61% expressed concern that health plan AI is increasing denial rates. Physicians reported spending an average of 14 hours per week on PA-related administrative tasks. The survey also found that 94% of physicians reported PA causes care delays and 93% reported poor clinical outcomes directly tied to PA burden. Read the findings.

Key Finding: Prior authorization is not a paperwork problem. At the rates physicians report, it is a patient safety problem.

Clinical Implication: Document PA-related delays in your patients' charts when they occur. This creates a medical record of harm that supports both appeals and future policy reform. It is also the right clinical practice.

KFF Health Policy, 2026

KFF's comprehensive analysis of federal and state consumer protections around AI in PA and claims review found that four states passed laws in 2025 explicitly prohibiting AI as the sole basis for a coverage denial, and that multiple additional states have active legislation in 2026. The analysis identifies a significant regulatory gap: even where human review is required, no federal standard specifies what that review must entail, how long it must take, or what qualifications the reviewer must have. Read the analysis.

Key Finding: State-level protections are growing but inconsistent, and the White House's March 2026 preemption framework threatens to erase them before they can take effect.

Clinical Implication: Know your state's current law on AI-based PA denials. If you practice in a state with protections, cite them when appealing. If you do not, the AMA's model state legislation is a reference point for advocacy with your state medical association.

The White House's March 2026 AI policy framework includes legislative recommendations to preempt state AI laws it characterizes as imposing "undue burdens" on AI development — a category that legal analysts note could encompass state laws banning AI-only PA denials, consent requirements for AI use in clinical settings, and other patient-protective regulations that have passed in states like Alabama, Indiana, and Utah. The framework is not binding and would require congressional action to take effect, but it signals a federal posture that is actively in tension with the state-level protections patients currently rely on.

Why This Matters: The same week a federal court ordered discovery into UnitedHealth's AI denial algorithm, the White House proposed limiting states' ability to regulate how that algorithm operates. Clinicians who care about the prior auth fight need to understand it is now also a federalism fight, and state medical associations are where that advocacy happens.

What It Does: Cohere Health is an AI-powered prior authorization platform designed to reduce the administrative friction on both the provider and payer side. For clinicians and their staff, it assembles the clinical documentation required for an authorization, matches it against current payer-specific criteria, and submits the request automatically. It is designed to identify missing documentation before submission rather than after denial, which is where most authorization delays actually originate.

Best For: Specialty practices with high PA volume — pain medicine, orthopedics, oncology, cardiology — where authorization burden is a significant share of administrative time. Also useful for larger outpatient clinics that want to reduce denial-driven revenue leakage without hiring additional staff.

One Practical Use Case: A pain medicine physician orders a spinal cord stimulator trial. Cohere Health pulls the patient's prior treatment history, maps it against the payer's current coverage criteria, flags any documentation gaps before submission, and submits the authorization with a complete clinical package. Average turnaround drops from weeks to days, and denial rates from incomplete documentation drop substantially.

One Limitation: Cohere is primarily a health plan and large health system tool; solo and small group practices may find the implementation overhead significant. It is not a plug-and-play consumer product — deployment requires integration with your EHR and payer contracts.

Prior authorization has always been a proxy fight — not really about whether a procedure is appropriate, but about who bears the cost and who has the leverage to make that determination. AI did not create that dynamic. It accelerated it, and it made the asymmetry between payer and provider more visible. When an algorithm with a 90% error rate on appeal can delay or deny care for millions of patients without a physician ever reviewing the chart, and when the appeals that reverse those denials require physician time that most practices cannot sustain at scale, the system is not malfunctioning. It is working as designed.

What is changing is the counter-pressure. Provider-side AI tools are narrowing the documentation gap that payers exploit. State laws are beginning to require human accountability for AI denials. Federal litigation is forcing discovery into how these algorithms actually work. None of it is fast enough for the patient waiting three weeks for a spinal cord stimulator trial while their pain goes uncontrolled. But the direction has shifted, and clinicians who understand the mechanics of the fight are better positioned to advocate for their patients within it.

The most useful thing any clinician can do right now is learn their payer's criteria for the procedures they order most, document PA-related delays in the medical record when they occur, and appeal every denial that is clinically unjustified. The data is clear: most AI-generated denials do not survive appeal. The payer is counting on the fact that most physicians do not have the time to fight.

Check out this Guide to help you get started using your own AI to collect data and prevent denials.

Send me an email letting me know what you’d like me to discuss in future issues @ [email protected]