Pangram is an AI-content detector for text and images, covering models such as ChatGPT, Gemini, Grok, Llama and Claude. Its outputs have featured in several disputes: it rated a Dartmouth provost's column 100% AI-generated, and was cited in accusations against author Thélyson Orélien and in reporting on Michelin's Dallas guide. NeurIPS used Pangram to desk-reject 178 submissions without human review, and independent checks later found high false-positive rates.
Michelin Guide accused of using AI to write Dallas reviewsD Magazine reported that most of Michelin's 2026 Dallas guide entries appear to be written by large language models, citing detection software and academic experts. Michelin denied using AI in its reviews, but Pangram and Carnegie Mellon researchers flagged flowery phrasing and grammar anomalies, raising doubts about the guide's credibility.
Dartmouth Provost Caught Using AI to Write ArticlesDartmouth provost Santiago Schnell was found to have used AI to write numerous articles, including an op-ed on student AI cheating. Detection tool Pangram judged his Aug 9 column 100% AI-generated; the college has opened an investigation and Schnell has apologized.
France in uproar over award-winning author accused of using AICanadian-Haitian author Thélyson Orélien, who just won the Prix du Roman Fnac, has been accused by anonymous group Balance ton Claude of writing his novel almost entirely with AI. Detection tool Pangram reportedly found 100% AI. Orélien and publisher Grasset deny it. The book is also shortlisted for the Goncourt and Renaudot prizes.
Pangram launches AI detector for text and imagesPangram released an AI detector for text and images, claiming high accuracy and third‑party verification from the University of Maryland and the University of Chicago; it detects content from models like ChatGPT, Gemini, Grok, Llama and Claude.
NeurIPS desk-rejected 178 papers using AI detectorThe NeurIPS Position Paper Track used a proprietary AI detector, Pangram, to desk-reject 18.4% (178) of submissions without human review. Independent checks found the detector flagged track chairs’ own papers at 24–69%, initially labeled nearly half of submissions as 90–100% AI, and was later retuned to reduce the flag rate to 12.7%.
The A.I. Daily