AI in Healthcare Prior Authorization: Why This Topic Is Suddenly Everywhere

Prior authorization has long been one of the most disliked processes in American medicine — and now artificial intelligence is running large parts of it, on both the payer side and, increasingly, inside CMS itself. Insurers say AI speeds up approvals and cuts waste. Physicians, patient advocates, and several state legislatures say it’s speeding up denials instead. Both things appear to be true, depending on the payer, the tool, and the year.

This piece lays out what’s actually documented: the federal rules now in force, the biggest lawsuit shaping the debate, what physicians report in their own surveys, and the state laws trying to put guardrails around algorithmic denials. For broader context on where AI helps and where it introduces risk across healthcare, see our related breakdown of real risks of AI in healthcare.

Quick Reference: AI in Healthcare Prior Authorization Developments

Development	What it does	Status
CMS Interoperability & Prior Authorization Final Rule (CMS-0057-F)	Requires payers to decide standard PA requests within 7 days, urgent ones within 72 hours, and give specific denial reasons	Effective Jan. 1, 2026, API rules by 2027
WISeR Model (Wasteful and Inappropriate Service Reduction)	CMS pilot using AI/ML plus clinician review to screen select Original Medicare services	Live in NJ, OH, OK, TX, AZ, WA from Jan. 2026
2026 CMS Interoperability Standards & Prior Authorization for Drugs (proposed)	Extends electronic PA requirements to drugs, adds API reporting	Proposed rule, comment period through June 15, 2026
California SB 1120 (“Physicians Make Decisions Act”)	Bars AI from autonomously denying, delaying, or modifying PA requests; requires licensed clinician sign-off	Effective Jan. 1, 2025
Lokken v. UnitedHealth Group (nH Predict lawsuit)	Class action alleging AI tool overrode physician judgment on post-acute care with a high error rate	Ongoing, in discovery as of March 2026

Why Insurers Are Turning to AI for Prior Authorization

The efficiency case is real, at least on paper. UnitedHealth reports that its Optum Real platform processes insurance eligibility and coverage determinations through real-time API connections.A separate pharmacy tool, PreCheck Prior Authorization, has compressed Rx approval time from more than eight hours to under 30 seconds, and nearly 95% of prior auth requests are now submitted electronically, with roughly half resolved in real time. That kind of throughput is the pitch every payer using AI for prior auth makes: fewer phone trees, fewer faxes, faster answers for routine, low-risk requests.

But faster isn’t the same as fairer, and that’s where the debate gets sharper. This is closely related to a theme we’ve covered before , see our piece on operational efficiency in healthcare vs. clinical AI, which looks at where automation genuinely helps operations versus where it creates new clinical risk.

The Central Risk: AI-Driven Denials and the nH Predict Lawsuit

The case that put AI prior authorization on the national radar involves UnitedHealthcare, its Optum subsidiary naviHealth, and a tool called nH Predict. The algorithm, built on a database of 6 million patient records, generated a predicted length of stay for elderly Medicare Advantage members receiving post-acute rehabilitation care, and according to the lawsuit, UnitedHealth set internal targets requiring clinical staff to discharge patients the moment the algorithm said they should no longer need care — regardless of what the treating physician determined was medically necessary.

The numbers cited in the complaint are stark: plaintiffs allege nH Predict has a 90% error rate due to lack of human review in the coverage denial process, with over 80% of the prior authorization denials being reversed on appeal. A federal judge has allowed the case to move forward, and in March 2026 a magistrate judge in Minnesota ordered UnitedHealth to produce a wide range of documents, largely siding with plaintiffs on six of seven discovery categories. Optum disputes the core allegation directly: “Claims that naviHealth is used to make adverse benefit or coverage decisions are false,” an Optum spokesperson told Becker’s.

There’s independent data suggesting the pattern predates the lawsuit’s specific claims. A 2024 report by the U.S. Senate Permanent Subcommittee on Investigations found that UnitedHealth’s prior authorization denial rate for post-acute care rose from 10.9% in 2020 to 22.7% in 2022, a period that coincided with the company’s implementation of AI-assisted authorization tools. Correlation isn’t proof of causation, but it’s exactly the kind of divergence regulators and plaintiffs’ attorneys are now scrutinizing.

What Physicians Report: The AMA’s 2025 Prior Authorization Survey

Beyond any single lawsuit, the American Medical Association’s annual survey gives the clearest national picture of how prior authorization — AI-assisted or not — is landing on frontline physicians. Based on 1,000 practicing physicians surveyed in December 2025, the findings are blunt: physicians complete an average of 40 prior authorizations per week, and nearly one in three (32%) report that requests are often or always denied. Nearly a quarter of physicians (26%) reported that prior authorization led to an adverse event for a patient, and more than 9 in 10 reported that prior authorization negatively impacts patient outcomes (92%) and delays access to care (95%).

Trust in insurer promises is also thin. In June 2025, roughly 60 insurers pledged reforms, yet as part of that pledge insurers committed to ensuring all medical necessity denials would be reviewed by a licensed and qualified clinician — yet only one in four physicians (24%) report that such reviews are consistently conducted by appropriately qualified clinicians. Among insurers named, three-quarters of physicians who work with UnitedHealthcare say the prior authorization burden in their practice is “high” or “extremely high” — the worst mark of any major commercial health insurer in the survey. For a broader look at how these pressures show up across care settings, our guide to real-world applications of AI in healthcare covers where automation is landing well versus where it’s controversial.

New Federal Rules Governing AI in Healthcare Prior Authorization

The CMS Interoperability and Prior Authorization Final Rule (CMS-0057-F)

This is the biggest structural change on the payer side. The rule enhances policies from the 2020 CMS Interoperability and Patient Access final rule and adds new provisions to increase data sharing and reduce payer, provider, and patient burden through improvements to prior authorization and data exchange practices; impacted payers were required to implement certain provisions by January 1, 2026, with API requirements due primarily by January 1, 2027. Practically, that means payers must now decide standard PA requests within 7 calendar days, decide urgent requests within 72 hours, and can no longer hand providers a generic denial — they must state exactly why the request was denied.

The WISeR Model: AI Screening Comes to Original Medicare

Original Medicare has traditionally avoided prior authorization altogether, but that’s changing. A new pilot program introduced prior authorizations for certain Original Medicare services in six states, effective January 2026, with proponents saying it will reduce wasteful spending by leveraging AI to expedite the process — while critics express concern about limits on patient access and added provider burden. Importantly, CMS says the pilot program will use AI but will not completely replace review by human clinicians. The model currently runs in New Jersey, Ohio, Oklahoma, Texas, Arizona, and Washington.

State Laws: California’s “Physicians Make Decisions Act” and Beyond

While CMS pushes AI into Medicare, states have moved to restrain how insurers use it commercially. California’s SB 1120, effective January 1, 2025, is the most detailed example: it explicitly prohibits the autonomous use of AI or algorithmic tools in making prior authorization denials, requires final determinations to be made by licensed clinicians, and requires plans to disclose AI use in utilization review and conduct periodic audits of those tools. As bill author Sen. Josh Becker put it, “The new law ensures that the human element will always determine quality medical treatments for patients.”

California isn’t alone. As of 2026, 37 states have enacted or introduced legislation governing AI in healthcare; Texas SB 815 prohibits AI as the sole basis for adverse determinations, and Maryland, Nebraska, Alabama, Indiana, Arizona, and Washington have enacted similar protections requiring human physician oversight. Coverage and enforcement still vary widely by state, so what’s true in California is not automatically true in a neighboring state.

What This Means for Clinicians and Health IT Teams

None of this points to a simple verdict on AI in prior authorization. The technology genuinely compresses timelines for routine, well-documented requests — the sub-30-second pharmacy approvals UnitedHealth describes are a real operational win when they work as intended. But the nH Predict litigation and the AMA’s own physician data suggest the same automation can, without careful guardrails, substitute population averages for individual clinical judgment — precisely the failure mode CMS’s WISeR model and states like California are now trying to legislate against.

For hospital leaders and health IT teams, the practical takeaway is to treat AI-assisted prior authorization the way you’d treat any high-stakes clinical decision support tool: know which vendor tools your payers use, ask how denials are audited against physician-only benchmarks, and track your own denial and appeal-reversal rates by payer. If you’re building internal governance around AI tools generally, our guide to AI in healthcare pros and cons is a useful starting framework for balancing efficiency gains against patient-safety risk.

FAQ

Does Medicare use AI to approve or deny prior authorization requests?

Original Medicare has historically not required prior authorization, but CMS’s WISeR Model pilot uses AI but will not completely replace review by human clinicians, seeking to speed up a process that can be lengthy and burdensome for patients. It currently applies only to select services in six pilot states, not to Medicare broadly.

Can an insurer use AI to deny a prior authorization request without a doctor reviewing it?

In California, no — SB 1120 explicitly prohibits the autonomous use of AI or algorithmic tools in making prior authorization denials, and requires final determinations to be made by licensed clinicians. Similar protections now exist in states like Texas, Maryland, and Arizona, but rules vary and not every state has such a requirement.

Why are so many AI-related prior authorization denials overturned on appeal?

In the UnitedHealth nH Predict case, plaintiffs allege the tool has a 90% error rate due to lack of human review, with over 80% of denials being reversed on appeal. Separately, appeal rates may understate the true error rate because, per one analysis, only a small portion of denied prior authorization requests were appealed in 2024, even though 80% of appeals are partially or fully overturned.

What is the CMS WISeR Model and which states does it cover?

WISeR stands for Wasteful and Inappropriate Service Reduction — a CMS Innovation Center model that leverages AI and machine learning, combined with clinician review, to ensure select items and services are furnished in line with Medicare coverage criteria. It launched January 2026 in New Jersey, Ohio, Oklahoma, Texas, Arizona, and Washington.