Pangram misses most Muse-Glimmer rewrites of scientific abstracts, Tokyo study finds

A Tokyo Metropolitan University paper finds that the AI-text detector Pangram missed 79.8% of scientific abstracts rewritten by Meta's Muse-Glimmer model, according to a report by Rohan Paul.

The same paper found Pangram caught 93.5% of abstracts rewritten by GPT-5, so the detector's miss rate depends strongly on which LLM produced the rewrite. Pangram flagged only 1 of 5,000 human-written abstracts, per the same report.

The Rohan Paul report is a social media post summarizing the paper, and the figures come from his account of it rather than from the paper itself.