Courts and regulators are probing AI in claims
A court allowed discovery into an insurer’s use of AI to deny claims, and legal scrutiny is expanding as regulators question whether AI-driven decisions lack proper human oversight — plus communications with generative AI may not be privileged. That legal appetite makes auditability and human-in-the-loop controls a sales-table requirement for automated adjudication. ( )
Estate of Gene B. Lokken v. UnitedHealth Group, No. 23‑CV‑3514 (JRT/SGE), is the Minnesota federal case at issue and the magistrate judge signed an order on March 9, 2026 directing discovery into UnitedHealth’s use of the nH Predict algorithm developed by naviHealth. (cases.justia.com (cases.justia.com)) Plaintiffs asked for seven discrete categories of documents — policies/procedures for post‑acute care, materials on the development and use of nH Predict, business and financial data tied to nH Predict, internal and government investigations, naviHealth employee incentive plans, oversight by UnitedHealth, and identities of employees issuing Notices of Medicare Non‑Coverage. (natlawreview.com (natlawreview.com)) The court’s March 9 order granted the plaintiffs’ second (development/use) and sixth (oversight) requests in full, denied the third (acquisition/financial benefit) and fifth (employee incentive) requests, and ordered partial production including employee training materials, government‑investigation records (but not purely internal probes), and limited employee identities. (cases.justia.com (cases.justia.com)) Judge John R. Tunheim had earlier ruled on February 13, 2025 that plaintiffs’ breach‑of‑contract and breach‑of‑good‑faith claims could proceed (while other state law claims were preempted), confining the litigation to whether UnitedHealth’s Evidence of Coverage promised human clinical decision‑makers. (govinfo.gov (govinfo.gov)) In a separate landmark ruling, Senior Judge Jed S. Rakoff of the Southern District of New York held on February 17, 2026 in United States v. Heppner that written exchanges with a publicly available generative AI platform (Anthropic’s Claude) were not protected by attorney‑client privilege or the work‑product doctrine. (debevoise.com (debevoise.com)) Regulatory pressure is converging with the courts: the NAIC’s Model Bulletin requires insurers to maintain a written AI program covering governance, vendor management, consumer notice and auditability, and as of March 2025 roughly 24 states had adopted the Model Bulletin or similar guidance. (content.naic.org (content.naic.org)) (quarles.com (quarles.com)) Industry advisory pieces and regulator‑facing commentary published March 24, 2026 say examiners are now asking insurers “Can you demonstrate how your AI models are governed?” and urge insurers to produce governance documentation, model monitoring logs, and human‑review protocols during examinations. (plantemoran.com (plantemoran.com)) (fenwick.com (fenwick.com))
