kapynResearch

AI text detectors struggle when language models mimic an author's style

Leading AI text detectors fail to reliably identify AI-generated content when it mimics an author's style. Epoch AI found that up to 18% of such AI-generated passages, and a concerning 48% in scientific writing, were misclassified as human-written. This highlights a critical challenge for academic integrity and content authenticity as LLMs become more sophisticated.

The Decoder·Jul 19, 2026

Opening Kapyn…