Medical AI Can Pass The Test And Still Fail Patients - OpEd - Eurasia Review

New reasoning models still reproduce medical stereotypes. Safety evaluation should expose subgroup failures before aggregate performance hides ...