Court allows broad discovery into insurer's AI coverage decisions

In Estate of Gene B. Lokken v. UnitedHealth Group, Inc., 2026 WL 658883 (D. Minn. 2026), the United States District Court for the District of Minnesota examined the boundaries of discovery in a lawsuit challenging an insurer's use of artificial intelligence in Medicare Advantage coverage decisions. The plaintiffs alleged that UnitedHealth's deployment of an AI tool called nH Predict was inconsistent with plan materials that explicitly stated coverage decisions would be made by clinicians. Because this allegation centered on whether the insurer honored its contractual commitments, the court granted discovery into a broad range of operational documents: policies, procedures, governance structures, regulatory oversight, training materials, oversight mechanisms, and vendor relationships — essentially everything that would show how the AI tool functioned in the real world.

At the same time, the court imposed meaningful limits. It denied plaintiffs' requests for certain internal investigations and financial or profitability data that were not sufficiently tied to the contract claims. More significantly, it rejected requests for the AI system's source code, underlying data, and embedded medical guidelines, recognizing that insurers have legitimate interests in protecting proprietary technical information. This distinction — allowing discovery of governance and operational detail while shielding technical architecture — reflects a calibrated approach: the court wanted to test whether the insurer lived up to its promises about human clinical involvement, not to expose the mathematical internals of the algorithm itself.

The court extracted three practical lessons for insurers. First, using AI in claims handling opens the door to broad discovery about implementation, supervision, and real-world use — especially when plaintiffs argue that AI has replaced individualized clinical judgment. Insurers must be able to document and demonstrate that AI serves as a support tool rather than a substitute for required human review. Clear documentation of how AI recommendations are generated, reviewed, and acted upon becomes essential. Second, the language in policy documents and evidence-of-coverage materials is critical. The court's entire analysis hinged on whether UnitedHealth's actual workflow matched its representations about how coverage would be decided. Any gap between what the documents promise and what happens in practice exposes the insurer to litigation and discovery risk. Third, proprietary AI materials may receive protection, but operational and governance-level information will not. Courts will likely shield source code and technical data while still requiring production of documents showing how AI affects claims handling, oversight, and decision-making.

The Lokken decision underscores that courts may scrutinize insurers' use of AI in claims handling and permit broad discovery into operational practices when contractual compliance is at issue. The ruling creates concrete incentives for insurers to align policy language with real-world workflows, maintain robust oversight records, and clearly document the role of human review in AI-assisted decisions — steps that can substantially reduce both litigation risk and the scope of discovery exposure if a claim is filed.

The Lokken decision reflects an important tension in how courts balance innovation protection against transparency when AI intersects with contractual obligations. The court's core holding turns on the specificity of UnitedHealth's policy language: because the plan materials promised that coverage decisions would be made by clinicians, the plaintiffs had a plausible claim that the AI tool's deployment violated that promise. This gave them a legitimate reason to probe the insurer's actual workflows, governance, and human oversight — not to reverse-engineer the AI, but to verify whether the contractual obligation was met in practice.

What makes Lokken significant for the insurance industry is that it narrows the scope of "trade secret" protection in AI claims. The court protected code and technical algorithms — the hard intellectual property — but not the operational and governance layers that bridge between algorithm and coverage denial. This distinction reflects a practical recognition: the question of whether a human clinician actually reviewed a denial is not proprietary in the same way the algorithm's weights are. It is a question of process compliance.

Insurers operating under this precedent should read it as a warning that vague or aspirational language about human involvement in their policy documents will create discovery exposure if their actual processes do not match. Clear documentation of oversight, training, and decision-making workflows becomes not just a compliance best practice but a litigation liability hedge. The case suggests that courts will scrutinize AI use in claims handling when contractual promises are involved — and that scrutiny will extend to how the tool is governed and applied, even if the tool's technical internals remain protected.

No comments yet. Be the first to share your thoughts!