Bestselling novel pulled from French literary prize after AI accusations — what next?

The timeline of events is not complicated: an anonymous X account accused Haitian-Canadian writer Thélyson Orélien's novel C'était ça ou mourir of being "almost entirely written by artificial intelligence"; subsequently, the Goncourt Academy withdrew the book from the Goncourt Prize shortlist, citing "the consistent conclusions of analyses by credible researchers and journalists." The work had previously topped France's bestseller list, sold roughly 35,000 copies, and won France's Fnac Novel Prize last Monday, receiving widespread acclaim.

But "withdrawal from the shortlist" is a confirmed institutional action, while "the novel was written by AI" remains an accusation. What the Goncourt Academy cited were analytical conclusions, without disclosing the specific methods and evidence; the anonymous account's accusation itself does not constitute a factual determination. Keeping these two layers distinct is the premise for understanding the entire controversy.

The accusation was not entirely anonymous to the end. Le Figaro columnist Samuel Fitoussi admitted to participating in the "group project" operating the X account, but denied political motives, saying he was defending "the moral contract between authors and readers" and hoping to raise awareness about AI writing. This admission shifted the event from "anonymous cyberbullying" to "an accusation campaign with specific participants and a public rationale."

However, the full membership and motives of this "group project" have not been fully disclosed. Fitoussi's self-explanation is a statement from a party involved, and cannot be equated with proof of the accusation's truthfulness.

Orélien denies using AI to write. He says he has suffered "a great deal of racist attacks" and is compiling materials with his French and Quebec publishers to prove his claim. The timeline his French publisher provides is: he began writing the novel in 2017 and completed the draft in 2019, long before generative AI became commercially available.

This timeline comes from the publisher and is a party's claim, not independently verified fact. Likewise, the proof dossier Orélien promised to compile has not yet been published, and its contents and effectiveness are unknown. Until the dossier is made public, both sides remain in a state of "each saying their own piece."

The detection tool used to make the accusation, Pangram, is called the "gold standard" of AI detectors. Research shows Pangram's accuracy in detecting AI exceeds 95%, with almost no false positives—but false positives still occur. This "almost" is precisely the crack in the controversy: in a high-stakes scenario like literary authentication, even an extremely low misjudgment rate is enough to destroy an author's reputation.

Pangram is not the first time it has been embroiled in such controversy. It has also been used to accuse other recent literary works, including Jamir Nazir's Commonwealth Writers' Prize-winning short story, which was eventually formally cleared. This precedent shows that accusations from detection tools can trigger serious consequences, while vindication often comes later.

French AI expert Benoît Raphaël conducted a set of tests: he input chapters from two, later four, Haitian novels into Pangram, and all were correctly identified as human-written; he then input Orélien's poems, short prose and reviews published before 2023, and these too were identified as human-written. These results point to a possibility: Pangram may not systematically misjudge non-standard, highly colloquial French texts.

But Raphaël himself is not a neutral skeptic. He says that before the scandal broke, he had written to another AI expert, suspecting Orélien's novel was written with AI assistance, and pointed out that its prose has typical signs of AI-generated writing, such as over-reliance on certain rhetorical structures. His test samples were chapters and pre-2023 works, not the full text of the novel in question. Therefore, this set of tests can neither prove the novel was written by AI nor prove it was not.

After the AI accusation, Quebec's La Presse accused Orélien of plagiarism in articles and reviews from 2012 to 2013; a Radio-Canada investigation found that at least five of his columns and reviews were partially copied from others' texts. These allegations and the AI writing claim are two different lines of evidence, but they reinforced each other in public opinion, together undermining the author's credibility.

It must be emphasized that these are media investigation conclusions, not judicial determinations. Orélien did not respond when contacted by La Presse. The existence of plagiarism allegations cannot automatically lead to the conclusion that "the novel was written by AI"; but it has indeed changed the public's receptiveness to the author's self-defense.

The French literary world's attitude toward this matter is not uniform. Writer Virginie Despentes publicly stated that Orélien has no need to prove himself, and that authors have the right to write in their own way. Raphaël Doan, who shared a stage with her and who in 2023 publicly used ChatGPT to create an alternative history work, said that despite Orélien's denial, he believes the author used AI, citing the novel's formulaic style, stale imagery and lack of plot coherence. Alexandre Gefen, editor-in-chief of A Cultural History of AI, compared Orélien to a Renaissance painter who had other artists in his workshop paint according to his instructions, arguing that authorship has always evolved with writing technology.

These disagreements point to the same unresolved question: the evidentiary status and industry standards of AI detection tools in literary authentication remain unclear. Whether Orélien actually used AI to write remains contested, with no final ruling. What is certain is that this controversy has turned "who is the author" from a private creative question into a public negotiation requiring institutions, tools and the public to adjudicate together.