In a study published in npj Digital Medicine on October 21, 2026, trainee clinicians accepted 52.0% of incorrect AI suggestions when reviewing radiographs with high-reliability AI models.
Three participating neonatologists in the Asan Medical Center study demonstrated sentinel behavior by correctly overriding 91.7% of AI errors when interpreting radiographs for neonatal pneumoperitoneum.
Researchers at Asan Medical Center in Seoul conducted a multi-reader crossover study involving 1,750 interpretation events for neonatal pneumoperitoneum to test clinician interaction with AI recommendations.
High-performing diagnostic AI systems may undermine the diagnostic skills of trainee clinicians by encouraging over-reliance on automated recommendations rather than developing adversarial resilience.
High AI reliability provided minimal incremental diagnostic gains to expert clinicians in the Asan Medical Center study, as experts were already performing at high accuracy levels.
Current regulatory frameworks for medical AI often require human-in-the-loop validation, but the Asan Medical Center study indicates that safety oversight depends heavily on clinician expertise.
The Asan Medical Center study deliberately exposed clinicians to both high-reliability AI models and systematically error-injected models to measure clinician vulnerability to automation bias.
Researchers at Asan Medical Center released an open-source educational sandbox designed to train clinicians against automated AI errors through deliberate practice.
The study involved five pediatric radiologists and two neonatologists at Asan Medical Center in Seoul, examining human-AI collaborative diagnosis in surgical emergencies.