Multiple independent evaluations reveal that while leading AI detectors can score above 90% accuracy on raw AI text, they often fail entirely against humanized AI content and produce high false ...