AI detector limitations and false positives
Detector outputs are probabilistic estimates. They are useful signals, but they are not legal-grade proof of authorship.
Where false positives appear
- Formal or institutional prose with constrained rhythm.
- Second-language writing with conservative phrasing.
- Highly edited text where stylistic variance has been reduced.
- Historical and canonical works that share modern detector features.
What a high score should trigger
- Manual review of pattern-level evidence.
- Check revision history and provenance.
- Request clarification before assigning blame.
What a high score should not trigger
Automatic punishment, public accusation, or irreversible decisions based on one detector result.
Evidence
Study known failures in the False Positive Hall of Fame and the full reliability brief: Do AI Detectors Work?.
Boundary
This project is anti-bypass. We publish transparent method guidance for better writing and better judgment, not for detector evasion.