Georgia Tech reframes medical AI liability across 31 US cases

Georgia Tech researchers Gennie Mansi and Mark Riedl, writing in Nature Health, analyzed 31 US legal cases involving AI harms to patients and concluded that the standard frame, treating physicians as the main line of accountability, misses where most of the harm actually happens.

The paper argues that "patients' medical care relies on a complex web of stakeholders" and that many AI tools deployed across that web negatively impact care. The systems named in the case pool sit outside the exam room: nH Predict, which has been at the centre of class-action litigation over Medicare Advantage denials; the Oregon DHS Safety at Screening Tool; and the Babylon Health Triage System.

Mansi and Riedl lay out two paths. One is "changing liability structures to reflect the role of many stakeholders in shaping how AI tools impact patient care." The other is "designing tools that can help advocates, such as patients' lawyers, provide critical legal expertise" and pursue recourse. The framing is explicit in the abstract: patients "have had no option but to seek legal recourse for harms," so the design brief shifts from the clinician who might refuse a recommendation to the lawyer who has to reconstruct the harm afterwards.

The cases were drawn from three databases: DAIL, the Health Litigation Tracker at Georgetown University Law, and AIAAIC. The abstract does not publish per-defendant outcomes or a breakdown of provider-side versus payer-side cases.