Full Breakdown
AI Authorship Allegations Cloud French Literary Prize for Thélyson Orélien
By Drooid · · How we work
The Allegations and Immediate Response
On September 21, hours after Thélyson Orélien received the Prix du Roman Fnac, an X account named *Balance ton Claude* posted screenshots claiming that multiple passages of his novel were identified as AI-generated by the Pangram detector, showing “100 percent AI” for the tested excerpts. The author denied the claim, stating he has never used AI and that the novel reflects his Haitian and Caribbean literary traditions. His French publisher Grasset described the accusations as “atrociously injurious” and affirmed the work’s originality. Canadian publisher Boréal also expressed confidence in the author’s work while acknowledging the allegations.
Background to the Controversy
The novel, published in France on August 19 by Grasset and in Canada in March by Boréal, tells the story of a Haitian teacher fleeing gang violence and seeking safety in Canada. It has sold roughly 35,000 copies in five weeks in France and is shortlisted for the Goncourt, Renaudot, Femina and Médicis prizes. The AI authorship claim revives earlier disputes such as the controversy surrounding Trinidadian writer Jamir Nazir’s Commonwealth Short Story Prize, which was temporarily withdrawn over similar allegations.
Detection Tool and Statistical Claims
The Pangram software, described by its developers as one of the most accurate AI-detection programmes, classified over 95 percent of the examined excerpts from Orélien’s novel as likely AI-generated, with a high confidence level. Independent studies cite a false-positive rate of 0.0041 percent on a million human-written English texts and no false positives in a sample of 495 human-written texts, though the developers acknowledge a non-zero error rate. Researchers at the University of Chicago caution that no detector can serve as sole proof of AI authorship.
Conflicting Reports and Gaps
Le Monde’s testing of more than half the novel produced the >95 percent AI classification, while other experts note that stylistic features such as rhythm, repetition and “living-word” imagery—characteristic of Orélien’s voice—might trigger false positives. Pangram’s performance varies by language, and passages taken from the same work are not statistically independent, undermining calculations that treat each test as a separate event. No public evidence, such as manuscript version histories or metadata, has been released to verify or refute the claim.
Official Statements & Responses
Orélien has asserted that the first draft was completed in 2019, before the widespread availability of generative AI tools, and is compiling documentation with his editors to demonstrate his authorship. Grasset condemned the accusations as malicious and maintained that the novel is an original creation. Boréal reiterated confidence in the author’s work. As of September 22, neither publisher had posted a detailed response on their official websites.
Broader Implications for Publishing
The episode highlights the difficulty of establishing authorship when detection tools provide probabilistic results rather than definitive proof. Scholars recommend supplementing AI-detectors with version histories, draft archives and metadata to create a more robust verification process. The controversy raises questions about how literary prizes and publishers will handle future works subject to similar scrutiny.
What Remains Unresolved
No independent audit of the manuscript’s digital provenance has been made public, and the authorship question remains open. The literary community awaits further documentation from the author and his publishers to clarify the writing process and address the reliability of AI-detection methods.