kapynResearch

AI chatbots reading X-rays can be dangerously confident even when they're wrong

AI radiology models exhibit dangerous overconfidence in incorrect diagnoses, according to the RadLE 2.0 benchmark. These models often present wrong findings with high certainty, underscoring the need for them to learn when to defer to human experts. Current AI lags behind human radiologists in diagnostic accuracy and appropriate confidence calibration.

The Decoder·Jul 19, 2026

Opening Kapyn…