AI Plagiarism Storm Engulfs France's Most Celebrated Debut Novel

A debut novel by Thélyson Orélien — a 38-year-old Haitian-Canadian author — has become the most high-profile AI writing scandal in literary history, threatening a remarkable rise that brought him Prix Goncourt consideration and publishing deals in more than 20 countries. The book, 'C'était ça ou Mourir' ('It Was Either That or Die'), tells the story of a Haitian teacher who flees gang violence and treks through South America to reach Canada. Released in March by Quebec's Éditions du Boréal, it won Fnac's novel of the year award and was set to be published in English by Scribner in 2028.

The scandal ignited from an anonymous X account called 'Balance ton Claude' — a riff on France's #MeToo hashtag — which claimed passages ran through the AI detection tool Pangram came back as 100% AI-generated. Academics then independently replicated the findings. Kevin Bryan, a University of Toronto professor specializing in AI economics, ran excerpts through Pangram and reported every selection returned a perfect AI score. Tuhin Chakrabarty, a Stony Brook University computer science professor who researches AI and publishing, fed the entire manuscript into Pangram and found 92% of the text flagged as machine-written. He stated flatly that no amount of argument could persuade him a 100%-flagged text was human-authored.

Researchers also pointed to specific syntactic patterns in the prose — constructions like 'Not A, Not B, but C' — as hallmarks of AI-generated text, particularly from Anthropic's Claude model. One revealing line from the novel's opening chapter, describing the protagonist's motive for fleeing Haiti, fits that template precisely. Bryan said he had no need for a detector tool to recognize the writing as AI-produced.

Orélien has firmly denied the allegations. In a statement, he said his style is rooted in Haitian and Caribbean oral traditions — poetry, repetition, and rhythm — and argued that reducing those literary techniques to evidence of AI authorship has nothing to do with genuine criticism. He says he completed his first draft in 2019, years before modern large language models became publicly available. His publisher, Éditions Grasset, backed him fully, condemning what it called malicious and peremptory claims. Philippe Claudel, president of the Académie Goncourt, also urged caution, noting that the initial accusation came from an anonymous account using unverified methodology.

Samuel Fitoussi, a writer and AI investor who identified himself as one of the people behind the Balance ton Claude account, framed the campaign as a defense of human writing in an era of AI-generated content. The accuracy of tools like Pangram remains contested broadly, but Pangram founder Max Spero cited internal testing showing a false-positive rate of just 1 in 24,000 for the platform's latest model. Critics have also raised the question of racial bias in AI detection, noting that three major AI writing scandals in the U.S. this year have targeted authors of color. Fitoussi and Spero both rejected that framing. With the Prix Goncourt announcement roughly a month away, the literary establishment faces an urgent and uncomfortable reckoning over whether its vetting processes are equipped for the AI era.