Tag
This paper evaluates commercial AI detectors in academic settings, finding high false-positive rates on AI-assisted human writing and near-total evasion via humanizers, concluding detector scores should not be standalone misconduct evidence.