AI text detectors are showing weaknesses when language models are prompted to imitate an author’s style. That makes automated detection less reliable in precisely the settings where the difference between assistance, imitation, and plagiarism matters most.
The issue is important for schools, publishers, and platforms that have looked to detectors as a scalable answer to generative writing. False confidence can harm legitimate writers while still missing sophisticated AI-assisted work.
The finding points toward a more cautious approach: provenance, disclosure, process evidence, and human review are likely to matter more than a single detector score.