Thélyson Orélien denies AI claims after French literary prize win

What happened

The French literary world is in turmoil over an accusation that a newly decorated novelist wrote his prize-winning book with artificial intelligence, BBC News reports.

Thélyson Orélien, a 38-year-old Canadian-Haitian writer, has just won the Prix du Roman Fnac — the first award of the French literary season — for his debut novel C'était ça ou mourir (It Was Either That or Die). He is also a finalist for the Goncourt and the Renaudot, the most prestigious annual prizes in French-language literature, which are due to be announced soon.

The accusation

According to the BBC, a little-known group called Balance ton Claude — a name that references Anthropic's Claude assistant and roughly translates as "denounce your Claude" — claimed on X that the novel was "almost totally" written by AI. The group said it ran passages taken from different parts of the manuscript through Pangram, a detection programme it called the most accurate available, and reported that every test came back "100 percent AI". Who belongs to the group is unclear.

The author's response

Orélien denied the charge immediately. Speaking to Libération, he said he had always loved literature and could not imagine why he would turn to a machine, and said he is assembling a dossier with his editor to demonstrate that he genuinely wrote the book. He pointed to his Haitian and Caribbean heritage, saying the novel's imagery, repetition and rhythm come from oral tradition and belong to him.

His French publisher Grasset also defended him, saying the accusations gravely harm a writer with an extraordinary life story and that the work is genuinely original.

Orélien was born in 1988 in Gonaives, in northern Haiti, wrote mostly poetry before emigrating, and later worked as a journalist in Canada, where he settled in Montreal in 2010. The novel follows a teacher who escapes gang violence in Haiti and eventually reaches Canada, where he regains, in the book's words, the right to breathe without fear. The BBC reports that his Canadian publisher calls it their biggest seller in 60 years, with roughly 20 translations planned. Orélien also told Libération that the first draft was written in 2019, before AI text generators became generally available.

How reliable are AI detectors?

The dispute hinges on whether such tools can be trusted. As the BBC notes, critics point out that detectors do not actually know who or what produced a text: they analyse statistical patterns typical of machine-generated writing, and those patterns can also occur in human prose. Different programmes often return different verdicts on the same text.

Research has found detectors can wrongly flag human writing and miss genuine AI output, particularly when the text has been edited or paraphrased. Some critics add that these tools are built around standardised English or French and struggle with dialect-inflected writing from places such as the Caribbean. The group behind the accusation counters that Pangram has been shown to misjudge a text only about once in every 24,400 cases.

The case echoes an earlier controversy the BBC highlights: Trinidadian writer Jamir Nazir's Commonwealth Short Story Prize was withdrawn in July over AI allegations, then restored after organisers said they were satisfied the writing was genuine.

Why it matters

The accusation lands just as France's literary season reaches its climax, with the Goncourt and Renaudot decisions pending. Whatever the outcome, the affair exposes a gap with no reliable way to close it: institutions are being asked to adjudicate authorship using tools whose accuracy — especially on writing outside standardised language norms — is contested. As AI-assisted drafting and editing becomes routine, there are still no settled norms for how much machine involvement is acceptable, leaving writers, and perhaps particularly those from marginalised linguistic traditions, exposed to verdicts delivered by detection software.