Court Orders Broad Discovery in UnitedHealth AI Suit

A federal magistrate judge in Minnesota ordered UnitedHealth Group on March 9, 2026, to produce a wide range of documents in a putative class action over its use of the nH Predict AI tool to deny post-acute care to Medicare Advantage members. The plaintiffs had moved to compel in seven categories, and the order reached every one of them, granting each at least in part while trimming specific requests inside them. It gives plaintiffs access to material that could show how the tool was developed, deployed, and overseen, with production due within 21 days.

The case was filed in November 2023 by the estates of two deceased Medicare Advantage members. The plaintiffs allege that UnitedHealthcare, through its naviHealth subsidiary, used nH Predict to override physician judgments about medically necessary skilled nursing facility care. An earlier ruling narrowed the case considerably: several of the state-law claims were held preempted by federal law, and what survives is breach of contract and breach of the implied covenant, limited to the terms of the plans’ evidence of coverage documents. UnitedHealth disputes the premise of the discovery it now has to produce, arguing that the requests rest on a “faulty premise” that nH Predict is used to make coverage determinations at all.

The date reach differs by request. UnitedHealth must produce its policies and procedures for post-acute care claims and employee training back to January 1, 2017, and performance, compensation, and discipline records for post-acute care coordinators and medical directors back to November 2017. The rest carries no start date of its own: internal analyses of nH Predict, records concerning the naviHealth acquisition and projected post-acute cost savings, government investigations into AI claims adjudication, and internal AI review board records along with the identities of its members. The order also reaches names and contact details for the medical directors and care coordinators who decided to issue non-coverage notices for 300 members of the proposed nationwide class. The court rejected UnitedHealth’s argument that documents predating the July 1, 2019 deployment of nH Predict were irrelevant, noting that 2017-2019 records could constitute circumstantial evidence, and pointing to a Senate investigation finding that the post-acute denial rate more than doubled after naviHealth and nH Predict came in.

The ruling matters for insurers beyond the specific allegations. It shows that courts are willing to order deep discovery into AI systems, including acquisition rationale, internal governance boards, and compensation incentives. Some requests were denied: nH Predict’s data, rules, source code, and underlying medical guidelines; internal investigations as opposed to government ones; blanket employee disciplinary records. There are limits to what plaintiffs can demand, but the breadth of what was granted is notable. The court also found that plaintiffs are entitled to discover how the tool works, its development goals and anticipated benefit, and whether it was designed to supplant physician decision-making.

Documentation written to satisfy an examiner now has to survive a deposition. AI tools used in utilization management or claims decisions need records built for that second audience. That includes evidence of clinical peer review, override mechanisms, and ongoing monitoring of denial rates and outcomes. It also means that internal communications about cost savings, productivity, and AI goals may become discoverable in future litigation. Carriers should review document retention policies and litigation hold procedures to ensure they can preserve and produce relevant materials without being caught by surprise.

The order also underscores the importance of the human-in-the-loop design. If a tool is described to regulators or members as supporting clinical decisions, the discovery process will test whether the design, training, and incentives actually support that claim. Carriers should be able to show that the AI is used to augment, not replace, clinical judgment, and that the final decision-maker has the authority and information to disagree with the AI.

The order arrives as several other AI denial cases move through courts and state legislatures tighten rules on AI use in health insurance decisions. The overlap between litigation and regulation means that carriers cannot rely on one defense alone; they must be able to demonstrate both regulatory compliance and fair process.

Discovery now runs ahead of the pleadings. The order reaches AI review board minutes and post-acute cost-savings analyses, so the case will test whether the internal account of how nH Predict was governed matches what was said to regulators and members, and whether those two agree is still open. Every health carrier running AI in utilization management should assume the same question reaches its own files eventually.