Maryland study: frontier models can unmask anonymous authors from a single text

A University of Maryland paper shows that with a single text and a zero-shot prompt, frontier models can determine whether text is human- or AI-written, identify the originating lab, and even name the author, with Opus 5.5 attributing 98% of AI texts to the correct lab.

2026-11-22 ~ 2026-11-22 · 2 related posts