The accuracy-safety trade-off of large reasoning models in clinical diagnostics
This study finds that reasoning-enabled AI models improve diagnostic accuracy but produce more false positives and less consistent outputs but physician oversight helps retain benefits while reducing safety risks.