Legal liability and AI in radiology: context and workflow matter

Abstract

The recent adoption of artificial intelligence (AI) in radiology raises important questions regarding radiologists' malpractice liability. While discussions often focus on algorithm performance, empirical evidence suggests that workflow design and contextual factors may be equally important. We conducted a series of experimental studies in which hypothetical jurors evaluated malpractice cases involving radiologists who used AI. We found that perceptions of liability were influenced not only by whether AI was correct or incorrect, but also by how AI was integrated into clinical decision-making. Providing AI error rates, documenting independent radiologist interpretations, and employing double-read workflows reduced perceived liability. These findings suggest that legal risk in AI-assisted radiology depends on clinical context, documentation practices, and workflow design, rather than algorithm accuracy alone.