Georgia Tech study: medical AI harms mostly bypass treating doctors

When AI harms a patient, the courtroom rarely finds a doctor at the source. A Nature Health analysis of 31 US legal cases and reported harms concludes that most complaints trace back to AI deployed by health insurers, state health departments and care facilities, not by treating physicians.

"Patients have had no option but to seek legal recourse for harms," write Georgia Tech researchers Gennie Mansi and Mark Riedl, arguing that accountability frameworks still treat physicians as the primary defendant even though "patients' medical care relies on a complex web of stakeholders."

The paper documents insurance algorithms terminating home-care coverage, a state screening tool that flagged a parent for child-welfare investigation, and triage systems that misdirected emergency care. In each of those settings, the AI decision that shaped what happened to the patient was made by a non-physician actor.

Mansi and Riedl argue for shifting from physician-centered to patient-centered accountability, both by rewriting liability structures to reflect the real chain of stakeholders and by designing explainable-AI systems that can help lawyers and advocates who represent patients after the fact. The preprint posted in July 2025; the Nature Health version appeared 21 September 2026.