Research shows educational institutes must not put too much faith in AI text detectors

A University of Florida study found five commercial AI text detectors to be highly unreliable on academic writing.

Tests on thousands of pre-ChatGPT papers and AI-rewritten text showed severe false positives, false negatives, and inconsistent accuracy.

Even simple paraphrasing caused major performance drops, suggesting detectors can be easily fooled.

The findings raise concerns that universities may be using flawed tools for high-stakes academic integrity decisions.