Nature study: AI harm suits skip treating physicians

An analysis in Nature Health of 31 US legal cases and documented harms involving medical AI finds that patients almost never sue their doctors. They sue insurers, state health departments, and care facilities whose algorithms overrode clinical judgment.

Gennie Mansi and Mark Riedl of Georgia Institute of Technology pulled the cases from three public databases in February 2025 and mapped who deployed each system. Named tools in the pool include UnitedHealth's nH Predict, which plaintiffs in Estate of Gene B. Lokken v. UnitedHealth Group say was used to deny care to elderly patients based on AI recovery predictions; Cigna's PxDx system, cited in Kisting-Leung v. Cigna and Van Pelt v. Cigna as accelerating claim denials; EviCore's algorithm, used to help insurers deny care; NarxCare, an opioid risk score; and a Veterans Affairs suicide-prevention algorithm. Physicians barely appear as defendants. "No harms or legal cases involved malpractice liability or claims in which physicians were suing or being sued for decisions made alongside AI systems," the paper reports.

The authors argue this exposes a misalignment in the explainable-AI research agenda, which has largely been built to help clinicians read model output. The determinations actually generating lawsuits are made by parties the patient's doctor never sees. "AI systems used by other stakeholders do not involve the patient's physician, but they make AI determinations that contradict physicians' clinical decision making," Mansi and Riedl write. They call for liability structures that reflect the "complex web of stakeholders" shaping care, and for XAI tools designed to support the advocates and legal representatives who pursue recourse for patients.