Unpopular Opinion: The Thélyson Orélien AI Accusation Is Junk Science

The Thélyson Orélien AI accusation turned the writer's best week into a public trial. The 38-year-old Haitian-born author, a former history-geography teacher living in Montreal, won the Prix du Roman Fnac on September 21 for his debut novel, C'était ça ou mourir, the book at the center of the storm.

The novel, published in France by Grasset on August 19, follows Jonas, a young Haitian teacher who flees gang violence and travels across the Americas to reach Canada. It had sold more than thirty-five thousand copies, with translation rights sold in over twenty countries, and it sat on the longlists for the Goncourt, the Femina and the Médicis. Scribner has confirmed plans to publish an English translation in the United States in 2028, according to the Wall Street Journal. That same day, an anonymous X account called Balance ton Claude posted results from the AI detector Pangram, claiming the novel was almost entirely written by a machine.

By Friday, the Académie Goncourt had removed the novel from its initial selection list, saying the move was driven by the desire to preserve the integrity of the prize. In the same week, posts on X and the Canadian newspaper La Presse surfaced claims that Orélien had committed acts of plagiarism in the early 2010s. The Thélyson Orélien AI accusation moved fast and had a single piece of technical evidence at its center: one anonymous account, one commercial detector, and a confidence figure its maker puts at 99.7 percent.

The machines cannot agree with each other

That confidence deserves a second look, because the detectors do not agree with each other. Agence France-Presse took the exact excerpts the anonymous account had flagged and ran them through InVID-WeVerify, a verification tool the news agency co-created. The result, according to reports on the testing, was that the passages were very likely written by a human. Other detection tools run on the same text returned varying results. So the technical case behind the Thélyson Orélien AI accusation is a single reading from a single commercial tool, contradicted by a tool built by one of the world's largest news agencies.

There is good reason to doubt these tools on prose like this in particular. Detectors are statistical guessers, and they are known to misfire on writing with unusual rhythms. Orélien's style is built on exactly that. "My traditions are Haitian and Caribbean. The images, the repetitions and the rhythms are the living word, they are my own," he said. A model trained to flag machine-like uniformity is a poor judge of prose that repeats on purpose, a reflex this column has criticized before, and the detector's defenders have offered no account of how it handles a Caribbean oral cadence translated into French literary form.

An accusation is not a review

None of this puts the book above scrutiny. The plagiarism claims deserve a serious look on their own terms, and prize juries have every right to investigate them. But the Thélyson Orélien AI accusation did its damage before any investigation finished. Orélien denied ever using AI tools to write, telling the French daily Libération that the attacks came as a surprise, even though he had faced racist attacks since arriving in France. "I am not sad, but angry," he said, adding that he was assembling a dossier with his publishers in France and Quebec to prove his case. Grasset condemned what it called malicious claims and reaffirmed its absolute confidence in him, and his Canadian publisher, Éditions du Boréal, said in a statement that it stands alongside him.

Wait for something indisputable

The unpopular opinion is this: the Thélyson Orélien AI accusation is not evidence, and a career should not be ended by a screenshot. Goncourt president Philippe Claudel said the academy would weigh the matter carefully if it had indisputable information by its October 6 meeting, and he called questions of AI authorship almost inevitable now. That is the right instinct, applied far too late. Claudel also called for a charter that would distinguish human-made work from machine-made work, and such a charter would be worth writing. Until it exists, the least the literary world can do is refuse to outsource its judgment to whichever black box shouts its confidence score the loudest. Tech panics usually say more about the panickers than about the technology, as this column argued about our AI relationship anxieties. Read the book, check the dossier, and keep the detectors where they belong: in the toolbox, not on the jury. You can read the full timeline of the dispute in this overview of the Orélien case, and Le Monde's analysis of what the controversy means for literature.