Artificial intelligence in radiology creates a paradoxical legal risk: when AI correctly identifies an abnormality a radiologist misses, juries are significantly more likely to hold the doctor liable, a Brown University study published in NEJM AI warns.
Lead author Michael Bernstein found what the authors call an “AI penalty.” In a simulated brain bleed scenario, mock jurors sided with the plaintiff 56 percent of the time without AI, but nearly 75 percent when AI had correctly detected the bleed and the radiologist disagreed.
“There is a real potential for legal repercussions if radiologists fail to find an abnormality that AI correctly finds,” Bernstein said. “It could be worse for them than if they fail to find something with no AI in the first place.”
The documentation dilemma cuts both ways. Medical malpractice attorney Rick Groffsky of Sommers Schwartz told Healthcare IT News that “when a clinician overrides a correct AI warning, the documentation now works against them.”
The study found one mitigating factor: when jurors were told the AI’s false omission rate, liability dropped from half to one-third. “People often think AI is magic,” Bernstein said. “When you make people aware it isn’t perfect, they have greater sympathy for radiologists who contradict it.”
For health systems, the finding underscores that AI governance must cover not just machine errors, but the documented trail AI leaves when it is right and clinicians are wrong.
