AI-writing accusation confronts French literary hit *C’était ça ou mourir*

Thélyson Orélien’s prize-listed novel *C’était ça ou mourir* has become the centre of a dispute over artificial intelligence after an anonymous X account alleged that it was written almost entirely by AI. The accusation, published on Sept. 21 by Balance ton Claude, concerned a book that had sold more than 35,000 copies and recently won the Prix du Roman Fnac. Balance ton Claude said it had run several passages through the detector Pangram and had “almost always” received a score of 100% AI. The account described Pangram as highly reliable, claiming a false-positive rate of 0.0041%, and said it had singled out the novel because of its commercial success. The allegations arrived as the French edition, published by Grasset on Aug. 19, appeared on the initial longlists for the Goncourt, Goncourt des lycéens, Femina, Médicis and Renaudot prizes. The novel follows Jonas, a young Haitian history and geography teacher who flees violence and insecurity in Haiti and travels across the Americas. One account of the story says he crosses 12 countries before reaching Canada. Grasset dismissed the claim as rumours and “fake news”. Orélien said he was assembling material to demonstrate that he wrote the book, and told the BBC that its first draft dated from 2019, before text-generating AI programmes were generally available. The author said that repetitions identified as alleged evidence of AI instead reflected a Haitian and Caribbean literary tradition. “People can argue about my style, but turning its devices into proof of artificial intelligence has nothing to do with literary criticism,” he said. He was born in Haiti and moved to Montreal in 2010. The book was first published in Quebec before Grasset acquired rights for French-speaking Europe. Orélien has also said he has faced numerous racist attacks since arriving in France, according to Euronews’ account of his comments to *Libération*. The dispute illustrates the limits of detector-based evidence. The BBC noted that such tools analyse linguistic probabilities and statistics rather than directly establishing authorship; they can wrongly flag human work or miss AI-generated text, particularly after editing or paraphrasing. Critics have also argued that systems trained on standard French or English may interpret Caribbean dialects poorly. Tests of the passages by different services have not produced a uniform result. AFP testing reported by BFMTV found divergent outputs from AI detectors, while InVID-WeVerify judged the selected excerpts very probably human-written. The Authors Guild likewise warns that detectors can generate both false positives and false negatives, putting authors’ contracts and reputations at risk. In its own test of 10 human-written articles published before 2023, Pangram returned a 0% AI score for every article. The guild says polished human prose can share statistical traits with machine-generated text because language models were trained on polished human writing. Benoît Raphaël said he had read the novel before the public controversy and believed it contained numerous AI markers; he said he sent the text to a friend on Aug. 24 after noticing them. Separately, *The Spectator* reported that a blogger’s Pangram test of 20 earlier Orélien texts, written between 2010 and 2017, classified 14 as fully human and six as mixed, with none classified as AI. The five people said to be behind Balance ton Claude later asked readers to ease attacks on Orélien and acknowledged a non-zero possibility that he had acted in good faith.