The 2026 Prix du Roman Fnac winner faces an AI authorship accusation that erupted within hours of his win, triggering a cascade of plagiarism allegations and a publisher suspension. Thélyson Orélien, the Haitian-Canadian author of C'était ça ou mourir, denies using artificial intelligence to write his bestselling debut, but conflicting detector results and the withdrawal of promotional support have placed the credibility of France's literary prize season under scrutiny.
The accusation and its timing
On September 21, the same day Orélien won the Prix du Roman Fnac, an anonymous X account with 92 followers posted results from the AI-detection tool Pangram. The account, Balance ton Claude, claimed it tested several passages and got a consistent result: "100% AI." The post concluded the novel was written "almost entirely" by artificial intelligence. It has since accumulated more than 4.9 million views.
Orélien's French publisher, Grasset, expressed its "absolute confidence" in him. His Canadian publisher, Éditions du Boréal, suspended all promotional activity for the book on Thursday, citing the author's well-being. The book remains on sale. Boréal called plagiarism "an unacceptable practice" and said it was "impossible at this stage to confirm or refute the allegations" regarding AI use.
Who is Thélyson Orélien and what is the novel about?
Orélien was born in 1988 in Gonaïves, Haiti. He studied law before leaving for Canada after the 2010 earthquake. He worked as a poet in Haiti and later as a journalist and columnist in Montreal. His writing is described as being "inhabited by memory, exile, and the question of belonging."
C'était ça ou mourir follows Jonas Dorléon, a teacher in Port-au-Prince whose neighborhood is stormed by armed teenagers. He escapes and crosses 12 countries - the Dominican Republic, Brazil, the Darién jungle, Central America, Mexico, and the US border - before settling in Canada. The 272-page novel has sold nearly 32,000 copies in France and is the biggest seller in Boréal's history since 1963. It is longlisted for the Goncourt, Médicis, Femina, Renaudot, and Prix Décembre.
Orélien's defense and the plagiarism allegations
Orélien said he began writing the novel in autumn 2017 and completed a first draft in 2019, before generative AI tools became widely available. "I have always loved creation and writing. I can't see why I would feel compelled to use a machine," he told Libération. He attributed the stylistic features critics flagged - repetition, rhythm, cascading imagery - to Haitian and Caribbean literary traditions, citing Aimé Césaire and Jacques Stephen Alexis as influences.
"I did not expect attacks of this kind. I am not sad, but angry," he said, adding that he had faced racist attacks since his recent arrival in France. By Thursday, he described a "will" to "silence" him.
On Wednesday, La Presse reported that a 2013 column Orélien wrote for the paper was almost entirely a lightly reworked translation of an article the Philippine Star published days earlier. Franceinfo identified another text containing a paragraph taken word for word from a 2014 opinion piece by former Colombian minister Rodrigo Botero Montoya. The plagiarism allegations concern journalism from more than a decade ago, not the novel.
What the detectors show - and don't
AI detectors look for statistical patterns, such as low "perplexity" or "burstiness," that correlate with machine-generated text. Pangram claims a false positive rate of about one in 24,400 documents based on its English-language benchmarks. Two recent studies have ranked it the most reliable detector on the market.
Results in Orélien's case point in conflicting directions. AFP ran the accuser's excerpts through its verification tool, InVID-WeVerify, which judged them "very probably written by a human." Seven other detectors AFP tested gave divergent results: one returned 100% AI, two concluded the opposite, and four varied by excerpt. La Presse submitted the entire novel to Pangram and got a score of 94%. CBC replicated the 100% result using different excerpts. Franceinfo's excerpts also scored 100% AI on Pangram, with one passage scoring 87% on GPTZero.
Radio-Canada ran the other Goncourt-listed novels available digitally in Quebec through Pangram, and all came back 100% human. Franceinfo tested Orélien's earlier work, published in 2014 and 2015, and Pangram classified it as 100% human - a finding that has given many pause, since it suggests the tool is not simply misreading his style.
A widening pattern of AI accusations
Orélien is the latest in a string of authors to face AI allegations this year. Nigerian author Jerry Falade lost a reported $2 million deal with Minotaur Books after his agents withdrew the manuscript, saying they could no longer verify how he wrote it. Mia Ballard, a Black American horror author, had Hachette pull her novel Shy Girl after a New York Times investigation. H M Wolfe faced accusations when Pangram scored her novel Daggermouth as 60% AI-generated; her publisher, Simon & Schuster, and her fans stood by her.
The Commonwealth Short Story Prize drew international attention when Pangram flagged "The Serpent in the Grove" by Trinidadian writer Jamir Nazir as AI-written. After reviewing drafts, timestamped documents, notes, and outlines, the Commonwealth Foundation cleared him and named him overall winner. Granta ended its decade-plus partnership publishing the prize's winning stories in the aftermath.
Falade has argued that all three Black authors who landed major deals this year had them cancelled or disrupted over AI suspicions. Many of Orélien's defenders see the same pattern in his case. Philippe Claudel, president of the Académie Goncourt, questioned the timing of the accusation: "You have an author who is Haitian and Canadian, who is Black, who is francophone, who is foreign, who writes about migrants." He said he had "at no moment" suspected the novel might have been written with AI and that he gives "the benefit of the doubt to those who are thrown to public opinion like this." The academy will announce its second shortlist on October 6 and said it will "obviously take a decision" if "irrefutable proof" emerges by then.
Why this matters for creatives, legal, PR and communications professionals, and writers
The Orélien case exposes a gap between AI-detection tools and the decisions publishers, prize committees, and communications teams must make under public pressure. Detectors produced contradictory results on the same text - some declared it definitively human, others definitively machine-generated. For legal and PR teams, the speed at which an anonymous account's claim became a reputational crisis is the operational lesson: Boréal suspended promotion within three days, and a literary festival cancelled a guest-of-honour appearance. For writers and editors, the case shows that a body of earlier human-written work may not protect against future accusations. The Goncourt's approach - demanding "irrefutable proof" before acting - offers one model for institutions weighing similar claims.
Your membership also unlocks:
