On March 9, 2026, a federal magistrate judge in Minnesota ordered UnitedHealth Group to turn over internal records detailing how its nH Predict algorithm was built and used, handing a significant discovery win to families suing the insurance giant over AI-driven coverage denials. The ruling, in Estate of Gene B. Lokken, et al. v. UnitedHealth Group, Inc. (Case No. 0:23-cv-03514, D. Minn.), is one of the most closely watched tests yet of whether an insurer can be held legally accountable for letting software, rather than a treating physician, decide when a patient’s care ends.
A 91-Year-Old’s Rehab Stay Becomes a Test Case
The named plaintiff, Gene Lokken, was a 91-year-old Wisconsin man who fractured his leg and ankle in the spring of 2022. His UnitedHealthcare Medicare Advantage plan covered roughly 19 days of rehabilitation care before, the lawsuit alleges, the nH Predict algorithm determined his stay should end, despite his treating providers believing he needed more time to recover. The case, filed in November 2023, has since grown into a proposed class action on behalf of other Medicare Advantage beneficiaries who say the same algorithm, developed by UnitedHealth subsidiary naviHealth, was used to cut off coverage for skilled nursing and rehab care.
What the March Discovery Order Actually Requires
Magistrate Judge Shannon Elkins ruled mostly in the plaintiffs’ favor on their motion to compel, ordering UnitedHealth to produce records showing how nH Predict was developed, what it was designed to do, who trained staff to use it, and whether it was built to override a physician’s own medical judgment. The order also covers documents about naviHealth’s internal AI Review Board and the cost-savings analysis behind UnitedHealth’s acquisition of naviHealth. Notably, the judge stopped short of compelling UnitedHealth to hand over the algorithm’s underlying source code, training data, or broad financial records, a limit that plaintiffs’ attorneys are expected to keep pressing on as the case moves toward later stages.
The Numbers Driving the Allegations
The lawsuit’s central statistic, drawn from a 2023 STAT News investigation, is that roughly 90% of nH Predict-driven denials that were appealed ended up reversed, a figure plaintiffs cite as evidence the algorithm was systematically wrong. Yet the same reporting found that only about 0.2% of policyholders ever file an appeal at all, meaning most denials, right or wrong, simply stand. UnitedHealth is not the only insurer facing this kind of scrutiny: Cigna has been sued over its PXDX system after ProPublica reported the company’s medical directors rejected more than 300,000 claims over two months without individually examining patient files, and Humana has faced similar claims over AI-assisted post-acute care denials. McKinsey has separately estimated that 50% to 75% of manual insurance-approval work is technically automatable, underscoring why insurers have leaned so heavily into these tools even as lawsuits mount.
UnitedHealth’s Defense: A Guide, Not a Gatekeeper
UnitedHealth and naviHealth have consistently argued that nH Predict does not itself make coverage determinations. A naviHealth spokesperson, Aaron Albright, has said the tool functions as “a guide to help inform providers, families, and caregivers” about the kind of care a patient may need, with actual coverage decisions based on CMS coverage criteria and the terms of each member’s plan. That framing matters legally: if the algorithm is merely advisory and a human reviewer signs off on every denial, UnitedHealth’s exposure narrows considerably. Plaintiffs’ attorneys counter that care coordinators were pressured to hew closely to the algorithm’s projected discharge dates, effectively making the software the real decision-maker in practice, which is exactly the question the newly compelled internal documents are expected to help resolve.
A Cautious Voice From Organized Medicine
The dispute has also drawn in physician groups wary of ceding clinical judgment to software. Dr. Marilyn Heine, a board member of the American Medical Association, has cautioned publicly that “AI is not a silver bullet” in utilization review, a position that reflects broader unease within organized medicine about insurers using predictive models to flag or terminate care without adequate physician oversight. That tension, between AI as an efficiency tool and AI as an unaccountable gatekeeper, is now playing out simultaneously in courtrooms and in state legislatures.
States Move Faster Than Washington
While federal regulation of AI in insurance remains limited, states have moved quickly. Alabama’s SB 63, signed into law in April 2026 and taking effect October 1, 2026, bars insurers from basing a medical-necessity denial solely on AI output, requires that any AI-assisted review account for a patient’s individual clinical history rather than group-level data alone, mandates periodic accuracy audits, and requires a qualified human professional to review adverse determinations. Texas enacted a similar restriction in 2025 barring automated systems from making adverse coverage determinations, and Becker’s Payer Issues has tracked at least seven states passing AI-specific health insurance laws in 2026 alone, a sign that statehouses, not Congress, are currently setting the rules of the road for algorithmic claims review.
The Lokken case is scheduled to proceed through further discovery in the coming months, with the newly compelled naviHealth and AI Review Board records expected to shape whether the dispute reaches a jury or settles beforehand. A related shareholder derivative lawsuit, reported in August 2026, separately alleges that UnitedHealth executives were aware of nH Predict’s high appeal-reversal rate well before the Lokken suit was filed, a claim that, if borne out in discovery, could sharpen the case for punitive damages. With Alabama’s law now in effect and more states weighing similar bills, insurers nationwide are watching whether courts and legislators converge on a single standard, or whether AI-assisted claims review ends up governed by a patchwork of state rules and ongoing litigation for years to come.
Photo: geralt / PIXABAY via Pixabay
