A new study from NYITCOM found that leading multimodal AI models made fundamental diagnostic errors in 20% of CT brain scan cases, with inconsistent explanations even among correct diagnoses.