Your Next Prior Authorization Decision May Be Made by AI. Should Practice Managers Be Worried?

Prior authorization has always been the most quietly hated word in a physician’s office. It is the fax that never arrives, the “additional information needed” letter that resets a two-week clock, the peer-to-peer call that never quite happens at the scheduled time. For the practice managers who own this process, it has meant hiring staff whose entire job is to fight a system that seems designed to wear people down.

Now that system is getting an upgrade — and it’s not the one anyone asked for.

Insurers are increasingly using artificial intelligence and automated decision-support tools in utilization review, at a pace that has outstripped both federal regulation and most practices’ internal workflows. The federal government itself has joined in: as of January 1, 2026, the Centers for Medicare & Medicaid Services (CMS) launched the Wasteful and Inappropriate Service Reduction (WISeR) Model in Original Medicare, using enhanced technologies including AI and machine learning alongside human clinical review. The pitch, from both payers and CMS, is speed and consistency. The fear, from physicians, researchers, and a growing bloc of state legislators, is that AI will not just approve care faster — it could also process denials faster, at scale, with less individualized human judgment behind each decision than ever before.

For practice managers, this isn’t an abstract policy debate. It is reshaping denial rates, appeal workload, staffing models, and cash flow in ways that are already measurable. Here is what the data actually shows, what’s changing in 2026, and what to do about it before it lands on your desk.

The Promise: Fewer Requests, Faster Answers

Insurers have genuine incentive to make the case that AI is fixing prior authorization, because the political and regulatory pressure on them has been intense. CMS Administrator Dr. Mehmet Oz has publicly pressed health plans to reduce the administrative burden of prior authorization.

There’s evidence of movement. An industry survey found that participating insurers had reduced prior authorization requirements by 11% between June 2025 and April 2026, representing millions fewer requests. The participating plans have also committed to standardizing electronic prior authorization submissions and reducing the number of services subject to review. On paper, the volume of prior authorization traffic is going down.

Individual payers are also publicizing efficiency numbers. UnitedHealth, which is investing roughly $1.5 billion in AI in 2026, reports a 96% first-pass approval rate through its AI-powered prior authorization technology. Humana has deployed Google Cloud’s Gemini-powered “Agent Assist” to more than 20,000 member advocates handling up to 80 million calls a year. Oscar Health says its AI agent, Oswell, now handles 86% of member inquiries, while its broader agentic-AI tools have cut peak-enrollment response times by 67%.

Read at face value, this looks like the industry finally solving its own worst problem.

The Catch: Nobody Actually Knows What’s Happening to Denials

Here is the sentence that should stop every practice manager mid-scroll: the same industry data showing an 11% drop in prior authorization requirements does not establish that denial rates have fallen by the same amount. Fewer requests is not the same as fewer refusals, and it is not the same as more appropriate refusals. A system can process fewer authorizations while denying a larger share of them — which is why denial and appeal data matter at least as much as request volume.

A team of Stanford researchers — writing in the wake of insurers’ rapid AI adoption — warned that AI risks amplifying the flaws that already existed in prior authorization, rather than correcting them, largely because of how little transparency and human review can sit behind AI-assisted decisions. Their concern isn’t hypothetical. Even before AI entered the picture, a substantial share of prior authorization denials didn’t survive scrutiny: studies have found high reversal rates on appeal, including an 82% overturn rate in Medicare Advantage plans. That number matters enormously once you put AI in the loop, because it suggests that at least some initial denial volume was already being reversed after human review — before AI-assisted systems were being deployed at greater scale.

A peer-reviewed perspective in npj Digital Medicine (February 2026) goes further, describing concerns about AI-assisted “blanket denials” of coverage, particularly in Medicare Advantage, and warning that opaque algorithmic systems can make it difficult to determine whether decisions reflect an individual patient’s circumstances.

The American Medical Association’s most recent survey of physicians puts a number on the anxiety: 61% said they are concerned that health plans’ use of AI is increasing prior authorization denials and exacerbating avoidable patient harms. The same AMA reporting cites Senate committee findings that some AI-driven review tools have produced denial rates as much as 16 times higher than typical baselines for certain services.

And this is the part that should live rent-free in every practice manager’s head: one of the most closely watched lawsuits in the broader AI-insurance space alleges that a major insurer used an AI model to evaluate and deny claims instead of relying on medical professionals, specifically to generate financial savings — allegations the insurer disputes and is contesting in court. Whatever the outcome, the case has become a reference point in the broader debate over AI-driven coverage decisions because it crystallizes the exact fear driving the policy response: a denial generated by a system whose clinical reasoning may not be transparent to the physician or patient.

What’s Actually New in 2026: Medicare Gets AI-assisted Prior Authorization for the First Time

The most consequential shift this year isn’t happening only inside commercial insurance — it’s also happening inside Original Medicare, which historically used prior authorization far more sparingly than Medicare Advantage.

CMS’s Wasteful and Inappropriate Service Reduction (WISeR) Model launched January 1, 2026, as a six-year model running through 2031 in six states: New Jersey, Ohio, Oklahoma, Texas, Arizona, and Washington. Under WISeR, CMS contracts with technology vendors to review a defined list of Part B services considered vulnerable to fraud, waste, and abuse or inappropriate use. The model uses enhanced technologies, including AI and machine learning, alongside human clinical review. Practices in affected states and specialties may be required to submit a prior authorization request before treatment or undergo a pre-payment medical review. (CMS)

A few operational details matter for anyone whose patients or referral base touches those six states:

Turnaround targets: CMS says standard determinations are expected within 72 hours, expedited cases within 48 hours.

Human clinical review is required. CMS says WISeR contractors must use clinicians in the review process; AI and other technologies are intended to support the review rather than replace clinical judgment.

Resubmission and peer-to-peer rights: Providers can resubmit after a non-affirmation and request a peer-to-peer discussion.

A “gold carding” exemption is planned. CMS has said it intends to test a feature that could exempt providers with consistently high approval histories from prior authorization or pre-payment review.

The politics around WISeR are unusually heated for a CMMI demonstration project. The model has drawn opposition from physician and hospital groups, including concerns that its use of technology in utilization review could delay or deny medically necessary care. The Society of Interventional Radiology, for example, has formally opposed the inclusion of vertebral augmentation procedures in the model.

In other words: this is not settled policy. It is a live fight, and the model is already operating while policymakers and provider organizations continue to challenge its design.

The Regulatory Backlash: States Are Moving Faster than Washington

While Congress argues over Medicare, state legislatures have become an increasingly important check on payer AI. By 2026, more than a dozen states had enacted laws or regulations addressing AI in health insurance coverage or utilization decisions, with seven states passing new AI health-insurance laws in 2026 alone.

Alabama’s SB 63 requires insurers using AI in coverage determinations to base decisions on the beneficiary’s medical history and individual clinical circumstances and includes disclosure and monitoring requirements. The law takes effect October 1, 2026.

Georgia’s SB 544 takes a different approach: it permits insurers to use AI for prior authorization and other functions, but prohibits AI from issuing an adverse determination without review and approval by a licensed healthcare provider. It takes effect January 1, 2027.

Iowa’s HF 2635 allows AI to be used for the initial review of prior authorization requests but prohibits it from being the sole basis for a decision to deny, delay, or downgrade medically necessary care. The law took effect July 1, 2026.

Washington’s SB 5395 prohibits carriers from relying solely on AI to deny, delay, or limit healthcare services through prior authorization and requires qualified human review of adverse medical-necessity determinations.

Minnesota’s HF 2500/SF 3984, by contrast, remained proposed rather than enacted, illustrating how quickly the legislative landscape can change from state to state.

At least 25 states have issued insurance-department guidance based on the National Association of Insurance Commissioners’ 2023 model AI bulletin, which applies across the insurance lifecycle, not just prior authorization.

Federally, CMS’s prior authorization transparency requirements took effect in 2026, requiring Medicare Advantage organizations, Medicaid and CHIP programs and managed-care plans, and ACA Marketplace issuers on the federal exchanges to publicly report specified prior authorization metrics, including approval and denial rates. The first reports, covering 2025 data, were due March 31, 2026.

The pattern across nearly every one of these laws is becoming clear: AI can assist, but it increasingly cannot be the sole basis for an adverse medical-necessity decision. That distinction is already emerging as a central principle in state-level regulation.

What this Means at the Practice Level

Strip away the policy debate and four practical realities are converging on practice managers right now.

1. Denial management needs to become proactive, not reactive. With high overturn rates on appealed Medicare Advantage denials, and AI-assisted systems now entering utilization review, a meaningful share of the denials your staff is fighting may still depend heavily on how the clinical record maps to payer criteria. That argues for tightening documentation before submission — ensuring clinical notes explicitly map to payer medical-necessity criteria — rather than relying on the appeal process to catch errors after the fact. Automated and AI-assisted review systems can be highly sensitive to documentation gaps and the way clinical information maps to payer criteria; closing those gaps up front reduces exposure regardless of what’s issuing the determination on the other end.

2. Track payer-specific AI and prior authorization transparency disclosures. The 2026 CMS reporting requirements mean Medicare Advantage, Medicaid and CHIP programs and managed-care plans, and applicable Marketplace issuers now have publicly available prior authorization metrics. That data is a genuine operational asset: it lets a practice see, plan by plan, which payers are denying at higher rates and adjust documentation strategy, staffing, or even payer contracting decisions accordingly, instead of guessing. The limitation is important: the initial federal data is aggregated and does not always reveal which specific services are driving denials.

3. Watch your state legislature, not just Washington. Because AI prior authorization regulation is developing state by state — and inconsistently, as the Alabama/Georgia contrast shows — a multi-state practice or health system may soon be operating under materially different rules on what a payer is even allowed to automate, depending on the applicable insurance market and state law. Build that into compliance tracking now, before it becomes a live billing dispute.

4. If you’re in a WISeR model state, understand the model’s provider protections and any future gold-carding process. For practices in New Jersey, Ohio, Oklahoma, Texas, Arizona, or Washington, understanding whether a service falls within WISeR and how non-affirmations, resubmissions, peer-to-peer review, and any future gold-carding mechanism work could materially reduce administrative disruption.

***

AI in prior authorization is not, on the current evidence, simply a story of insurers automating cruelty, nor is it simply a story of overdue efficiency finally arriving. It is both, moving at the same time, in the same systems, often for the same patients. Insurers have real efficiency numbers to point to. Physicians, researchers, and a growing list of state legislators have real reasons to worry that some automated systems can generate or contribute to denials without enough individualized clinical review.

The regulatory answer taking shape — AI can assist, but a human professional should remain responsible for an adverse medical-necessity decision — is increasingly reflected in state laws. Whether that principle becomes more uniform nationally is the question practice managers will be answering with their own denial logs, appeal outcomes, and payer disclosures for the next several years, not the one insurers will be answering in their press releases.

Sources