AI X-ray tools are too confident when they are wrong

Doctors use AI to help read X-rays, but these tools have a major problem. They often guess incorrectly and act like they are completely sure. This behavior is dangerous because it can mislead medical staff.

A new test called RadLE 2.0 measured how well AI models know when to step back and let a human take over. Most models failed to admit when they were uncertain. They kept making false claims even when the images did not support them.

Human radiologists are still much more reliable than current AI. For technology to play a bigger role in hospitals, it must learn to say I do not know. Until then, experts need to double check every AI result.