Georgia Tech study reframes medical AI liability in 31 US cases

The medical, legal and technology communities have centered physicians when they think about harms from medical AI tools. A Nature Health paper published 21 September 2026 by Georgia Tech's Gennie Mansi and Mark Riedl reads 31 US legal cases alongside reported harms and argues that framing is wrong.

"The medical, legal and technology communities often centre physicians' role in preventing and mitigating harms from medical artificial intelligence (AI) tools," the authors write. The case pool points elsewhere. Healthcare delivery, they say, runs through "a complex web of stakeholders—physicians, state health departments, health insurers and care facilities, among others," and accountability, on their reading, should follow that wider web.

The cases were assembled from three open databases as of February 2025: the Database of AI Litigation kept at George Washington University Law School, the O'Neill Institute litigation tracker, and the AIAAIC repository of AI incidents. Tools and incidents cited include the nH Predict system, an Oregon child abuse screening tool dropped over fairness concerns, and Medicare Advantage plans using AI to deny care for seniors.

The authors' bottom line: "patients have had no option but to seek legal recourse for harms." Their two proposals are a liability regime that reflects the multiple parties shaping how an AI tool lands on a patient, and "designing tools that can help advocates, such as patients' lawyers, provide critical legal expertise and practically support recourse for patients."