A review that didn’t settle the question
Éditions du Boréal’s decision to stop working with Thélyson Orélien turns a dispute over AI authorship into a question of trust between a writer and his publisher. The Montreal house says an outside review found AI markers in metadata from some working documents for Orélien’s novel, C’était ça ou mourir, though it can’t establish how much AI contributed. That distinction matters: the review raised questions about the files, but it didn’t determine who wrote the book’s prose or how any tools may have shaped it.
Published earlier this year, the novel follows a Haitian migrant who leaves for Canada, a journey that echoes Orélien’s own move after Haiti’s 2010 earthquake. The book received praise in French-language literary circles and won the Prix du roman Fnac on Sept. 21. The award brought new attention to the novel just as questions about its authorship were beginning to circulate.
That same day, an anonymous account on X posted results from Pangram, an AI-detection tool, which assessed excerpts as “100 per cent” generative AI. That result prompted scrutiny, but it wasn’t the basis of the publisher’s decision: Boréal says it commissioned a separate review of the novel’s working files. A detector’s score can attract attention, but the number alone doesn’t explain how a text was produced or establish what happened during the writing process.
Boréal says it spent several weeks working with Orélien to assemble documentation of his writing process, hoping to address the allegations. Those efforts failed to resolve whether AI had been used, the publisher said. Orélien later told the company he would provide no further information about how the novel was written. For a publisher trying to assess a contested manuscript, missing or incomplete process records can leave the central question open even after a review.
The publisher then commissioned an external analysis of all available working versions. Results delivered this week identified what Boréal described as traces of AI markers in the metadata of certain documents. The house says the evidence doesn’t reveal the extent of any AI use, and that it may never know. Metadata can offer clues about a document’s history, but the publisher’s own account stops short of saying those clues prove that AI generated passages in the finished novel.
Conflicting accounts and a break with the publisher
Orélien had repeatedly assured Boréal he didn’t use AI to create the novel. In a statement last week, he called the allegations “slander and heresy.” He acknowledged using AI tools “here and there” for research, while maintaining that the book’s text, structure and plot came “straight from my gut.” His account draws a line between using a tool during research and using one to produce the work itself; the publisher says its review couldn’t resolve the question that separates those claims.
He also admitted to separate plagiarism allegations, describing them as “unacceptable” mistakes from his youth. The publisher’s announcement doesn’t say those allegations drove its decision; it points to the unresolved AI question and the breakdown in confidence. Keeping those issues distinct matters, since the public discussion can easily collapse different allegations into one broad judgment about the book.
Boréal says it has ended its professional relationship with Orélien. It won’t reprint C’était ça ou mourir or publish any of his future books. The company says it informed the author before making the decision public. This is more significant than it looks: the dispute has moved beyond debate over a detector’s result and now affects the novel’s availability and the author’s relationship with its publisher.
The case leaves a gap between what AI-detection tools suggest and what the publisher says it could establish from the novel’s document history. Boréal says its role isn’t to investigate authors, but it considered further action necessary after the process review failed to resolve the dispute. The outcome shows the limits of a review when the available records raise concerns without answering how much, if any, AI shaped the published text.
What the dispute means for authors and publishers
For publishers, the case points to a practical problem: when authorship is challenged, the evidence may be incomplete, and the available tools may answer different questions. A detection score describes how a tool classified text; a document review examines files and their metadata. Neither, as Boréal describes its findings, settled the full question here. If you’re working in this space, the distinction between suspicion, evidence and proof will shape how disputes are handled.
And yet, decisions still have to be made. A publisher’s relationship with an author depends in part on confidence in the account of how a book was written, and Boréal says that confidence broke down. That’s a consequential judgment made amid uncertainty. Similar disputes may leave publishers weighing incomplete records against an author’s explanation, with no simple technical result that resolves the disagreement (and this is the part most people overlook).
As the controversy continues, the publisher has urged the public to refrain from personal attacks. “Beyond the book and the controversy, there is, first and foremost, a human being,” it said. The appeal recognizes a cost that technical arguments can obscure: public scrutiny can attach to a person even when the available evidence remains inconclusive.
Unresolved, and consequential.
