Georgia Tech: 31 cases show AI patient harms bypass doctors

Two Georgia Tech researchers reviewed 31 U.S. legal cases involving medical AI and found that most of the harms did not begin with a doctor at the bedside. They began with an insurer, a state agency, or a facility running a tool the treating physician often never configured.

The paper, published in Nature Health on 21 September 2026 by Gennie Mansi and Mark Riedl of the Georgia Institute of Technology, argues the law is pointed at the wrong actor. "patients' medical care relies on a complex web of stakeholders...and many AI tools deployed across their healthcare delivery negatively impact their care," the abstract states. In response, the authors write, "patients have had no option but to seek legal recourse for harms."

Reporting on the paper by Bioengineer picked out two of the fact patterns the study works through: Medicare Advantage plans using AI to deny post-acute care for seniors, and state screening tools that flagged parents for child welfare investigations.

The proposed fix has two parts. First, changing liability structures "to reflect the role of many stakeholders in shaping how AI tools impact patient care." Second, building tools that can help "advocates, such as patients' lawyers, provide critical legal expertise and practically support recourse for patients." That means logs, records and explanations a legal representative can actually use.

The abstract does not name any of the 31 cases or say which stakeholder appears most often; that detail sits deeper in the paper.