Operations Updated June 19, 2026

AI in Prior Authorization: What It Means for Diagnostics and Molecular Labs

Payers and providers are both deploying AI in utilization management and prior authorization, drawing new regulatory scrutiny in 2025-2026. Here is what the trend means specifically for diagnostics and molecular labs, and how to tell trustworthy automation from a black box.

Educational use only. This article is for general information and is not legal, compliance, billing, or coverage advice. The regulatory landscape around AI and prior authorization is changing quickly, and several items below are actively contested in 2026. Verify current law, regulations, and payer policy against primary sources and consult qualified professionals before acting. No outcome is guaranteed; use at your own risk.

A molecular oncology test gets ordered for a patient with suspected recurrence. The order is clean, the medical record supports it, and the test is on the lab's menu for a reason. Weeks later the remittance comes back as a denial: not medically necessary. No one at the plan called the ordering physician. There may not have been a human in the loop at all until an appeal forced one.

If that scenario feels familiar, you are living inside the central tension of prior authorization in 2026. Artificial intelligence is now on both sides of the table. Payers use it to screen, flag, and sometimes deny requests at scale. Providers and labs are racing to use it right back, to assemble documentation, predict denials, and appeal faster. And regulators, from Congress to state legislatures, are trying to draw a line between AI that makes the process faster and AI that simply makes "no" cheaper to produce.

For a diagnostics or molecular lab, this is not an abstract policy debate. Your revenue depends on coverage decisions that are increasingly machine-mediated, against policies that change constantly, for tests that are among the hardest in all of medicine to authorize. Here is what is actually happening, what is verified, and what to watch.

Physicians already believe payer AI is increasing denials

Start with how the people on the receiving end see it. In its 2024 prior authorization physician survey of 1,000 practicing physicians, the American Medical Association found that 61% of physicians are concerned that health plans' use of AI is increasing prior authorization denials, and 75% reported that the number of denials had increased somewhat or significantly over the prior five years (AMA, 2024).

That concern did not come from nowhere. In October 2024, the U.S. Senate Permanent Subcommittee on Investigations, then chaired by Senator Richard Blumenthal, released a report on Medicare Advantage prior authorization finding that major insurers were leaning on predictive technology to deny post-acute care. The report documented that in 2022, one insurer's denial rate for post-acute care was roughly 16 times higher than its overall denial rate (Senate PSI report, via AHCA/NCAL, 2024).

The point for a lab operator is not to assign blame to any one insurer. It is to recognize that automated utilization management is now an established part of how coverage decisions get made, and that the scrutiny it has attracted is producing real rules you will have to operate under.

The regulatory response is arriving in three layers

Layer one: CMS sets the clock and the plumbing

The most concrete federal change is the CMS Interoperability and Prior Authorization final rule, CMS-0057-F, issued January 17, 2024. For impacted payers, which include Medicare Advantage organizations, state Medicaid and CHIP fee-for-service programs and managed care plans, and most Qualified Health Plan issuers on the federally facilitated exchanges, the rule sets hard decision timeframes: expedited (urgent) requests must be answered within 72 hours and standard requests within 7 calendar days, generally beginning January 1, 2026. Payers must also give a specific reason for every denial. The FHIR-based API requirements, including a dedicated Prior Authorization API, generally carry a later compliance date of January 1, 2027 (CMS-0057-F fact sheet).

Two caveats matter for labs. First, traditional fee-for-service Medicare and most commercial plans are not impacted payers under this rule, so the timeframes do not uniformly cover your book of business. Second, faster decisions are only as good as the data behind them. A 7-day clock on a wrong answer is still a wrong answer; it just arrives sooner.

Layer two: states require a human behind the decision

While CMS standardized the timing, states have gone after the substance, specifically the question of whether a machine can deny care by itself. California's SB 1120, the Physicians Make Decisions Act, took effect January 1, 2025 and prohibits health plans and insurers from denying, delaying, or modifying care based solely on AI; medical-necessity decisions must rest on the enrollee's clinical history and individual circumstances and be reviewed by a qualified human (Fenwick analysis of SB 1120).

Maryland followed with a law effective October 1, 2025 that closely mirrors California's, requiring that AI tools support, rather than replace, clinician medical-necessity determinations based on the patient's full clinical picture, and adding reporting and audit obligations for carriers (Alston & Bird, 2025). A further Maryland measure, HB 1563, effective June 1, 2026, adds quarterly transparency reporting to the Insurance Commissioner on adverse decisions and whether AI was used (Holland & Knight, 2026).

These state laws generally reach state-regulated plans, not self-funded ERISA plans or Medicare, so their practical effect on a given claim depends on the plan type. But the direction of travel is unmistakable: a human clinician, not an algorithm alone, is increasingly required to own an adverse medical-necessity decision.

Layer three: the government tries AI itself, and Congress pushes back

The most fast-moving item is the CMS Innovation Center's WISeR Model, short for Wasteful and Inappropriate Service Reduction. It began January 1, 2026 as a six-year pilot running through December 31, 2031 in six states (Arizona, New Jersey, Ohio, Oklahoma, Texas, and Washington), testing AI-assisted prior authorization on a defined set of higher-cost Part B services such as skin and tissue substitutes, electrical nerve stimulator implants, and certain knee procedures. CMS has stated that AI may flag requests but that no claim will be denied without review by a qualified human clinician (Forvis Mazars, 2025).

WISeR drew immediate criticism over its incentive structure. According to the American Hospital Association, participating vendors stand to receive a share of the savings tied to reduced spending, which the AHA argued risks "a perverse incentive to deny care" even with a human-review requirement (AHA comment letter, 2025). The model is now under active repeal pressure. On May 12, 2026, the Government Accountability Office determined that the WISeR notice is a "rule" subject to the Congressional Review Act (GAO B-337994, 2026), and on May 20, 2026, members of Congress introduced a resolution of disapproval to repeal it, alongside separate legislation and an appropriations amendment aimed at blocking the pilot (Congress.gov, S.J.Res.192; Georgetown Medicare Policy Initiative, 2026).

Treat WISeR as genuinely unsettled. As of mid-June 2026 it is operating, but its scope and survival are contested, and what it tests today may not be what it tests, or whether it exists at all, a year from now. For labs, the more durable lesson is the precedent: even when a payer is the federal government, applying AI to coverage triggers fierce debate about who reviews the machine and who profits from a denial.

Why diagnostics and molecular labs sit at the sharp end of this

Most coverage of AI in prior authorization is written for hospitals and post-acute care. Diagnostics is a harder case, for structural reasons.

Molecular and genetic testing is documentation-dense in a way few other services are. Under Medicare's MolDX program, administered through Palmetto GBA, many tests must carry a registered Z-Code that is tied to a technical assessment of the test's analytical validity, clinical validity, and clinical utility, and payers increasingly use those identifiers to automate claim adjudication (XiFin, MolDX Z-Codes). When adjudication is automated, the match between the test performed, the exact policy in force, and the codes submitted has to be precise. A registered identifier that is slightly off, a policy that quietly updated, or a clinical-utility requirement the documentation does not address, and an automated "not medically necessary" follows.

Layer on top of this the pace of change. Molecular medical policies are revised frequently, coverage varies test by test and payer by payer, and new assays arrive faster than payer policy can settle. The same conditions that make automated payer review error-prone, ambiguous and shifting rules, make automated provider-side defense valuable, but only if the automation is reasoning over the policy that is actually in effect today.

The real dividing line: grounded automation versus the black box

Here is the strategic point, stated plainly. AI in prior authorization is neither savior nor villain. Its trustworthiness comes down to one question: what is it reasoning over?

A payer system that produces denials from an opaque model, with rules you cannot inspect and reasoning you cannot trace back to published policy, is the thing regulators in California, Maryland, and Congress are reacting against. The fix the new laws keep reaching for is the same: ground the decision in the individual's clinical facts, keep a qualified human accountable, and make the basis auditable.

The provider and lab side deserves the same standard. AI applied to reimbursement is only as good as the underlying payer rules it reasons over. If a tool generates a confident-sounding medical-necessity argument from stale, generic, or invented policy language, it is just a more articulate black box, and it will lose on appeal. The version worth trusting is grounded in real, currently published medical policy, with every assertion traceable to its source document, so that a human can verify it and a payer cannot wave it away.

That is the approach Converus takes: structured, current payer medical policy and coverage rules delivered through an API and MCP interface, so the AI tools and revenue-cycle systems your lab already uses can reason over real, source-linked policy instead of guesses. The goal is not to out-automate the payer. It is to make sure that when a decision, on either side, is challenged, the rule it rests on is one you can actually point to.

What to do with this now

Map which of your payers fall under CMS-0057-F's 2026 decision timeframes and which do not, because your follow-up cadence should differ. Track the AI-oversight laws in the states where your patients are covered, since they shape what a defensible appeal can demand. Watch WISeR's status if you operate in its six states, but do not build around a pilot that Congress is actively trying to repeal. And whatever automation you adopt, insist that it cite its sources. In a world where the other side's machine may not, traceability is not a nice-to-have. It is the whole advantage.

Sources

Frequently Asked Questions

Does CMS-0057-F require faster prior authorization decisions for labs?
Yes, for the payers it covers. Under the CMS Interoperability and Prior Authorization final rule (CMS-0057-F), impacted payers, which include Medicare Advantage, state Medicaid and CHIP fee-for-service and managed care, and most QHP issuers, must send expedited (urgent) prior authorization decisions within 72 hours and standard (non-urgent) decisions within 7 calendar days, generally beginning January 1, 2026. The required prior authorization and other FHIR-based APIs generally have a compliance date of January 1, 2027. Note that traditional fee-for-service Medicare and many commercial plans are outside this rule. (Source: CMS-0057-F fact sheet.)
Can an insurer use AI as the sole basis to deny my lab's claim on medical necessity grounds?
It depends on the state and plan. A growing number of states bar AI from being the sole basis for a medical-necessity denial. California's SB 1120 (effective January 1, 2025) and Maryland's law (effective October 1, 2025) both require that medical-necessity determinations rest on the patient's clinical information and individual circumstances, not an algorithm alone. These laws apply to state-regulated plans; self-funded ERISA plans and Medicare are governed separately. Always verify the specific plan and current state law.
What is the CMS WISeR Model and does it affect lab testing?
WISeR (Wasteful and Inappropriate Service Reduction) is a CMMI pilot that began January 1, 2026, testing AI-assisted prior authorization on selected services in Original Medicare across Arizona, New Jersey, Ohio, Oklahoma, Texas, and Washington. As announced, it targets specific Part B services such as skin substitutes, electrical nerve stimulators, and certain knee procedures, not broad lab panels. It is also facing an active repeal effort in Congress, so its scope and future are in flux. Verify current status before relying on it.
How is molecular and genetic test prior authorization different?
Molecular and genetic tests are unusually documentation-heavy. Under Medicare's MolDX program, many tests require a registered Z-Code tied to a technical assessment of analytical validity, clinical validity, and clinical utility, and payers increasingly use these identifiers to automate claim adjudication. That means a small mismatch between the test, the policy, and the codes submitted can trigger an automated denial, which is exactly where current, traceable payer-rule data matters most.
← Back to Knowledge Base