Thélyson Orélien's C'était ça ou mourir was doing everything a debut novel is supposed to do — 35,000 copies sold, the Prix Fnac, translation deals in more than 20 countries. Then, on the same day it won the Fnac prize, an anonymous X account called Balance ton Claude ran it through an AI-detection tool called Pangram and declared it 'almost entirely' machine-written. The Académie Goncourt, one of the most powerful literary institutions in Europe, eventually acted. The anonymous account was later identified as Samuel Fitoussi, a Le Figaro columnist whose previous published work is an essay against woke culture in film and television. The evidentiary foundation here is soft. Pangram flagged the novel; other AI-detection programs, tested by French media, found the same text 'very probably' written by a human. This is not a minor discrepancy — it is a fundamental problem with the entire category of AI-detection software, which produces false positives at rates that make them unusable as legal or institutional evidence. The Académie Goncourt's own president, Philippe Claudel, urged caution mid-week. Then came a separate plagiarism allegation from Libération, sourcing a short story Orélien submitted to a Canadian radio competition in 2012 to a Louis Pauwels preface from 1960, and the Academy changed course. Orélien has denied both allegations. He told Libération that the imagery, repetition, and rhythm of the novel are rooted in Haitian and Caribbean oral tradition — a living linguistic inheritance that, he says, AI-detection software almost certainly does not know how to read. This is not an implausible defense. Caribbean literary French, with its formal repetitions and lyrical refrains, can read as statistically unusual to tools trained on metropolitan French prose norms. The possibility that the tool flagged distinctiveness as artificiality is worth taking seriously — and the Académie did not appear to take it seriously at all. What the Academy's statement actually says is revealing. It bundles the plagiarism allegation (a distinct, older, and potentially more concrete charge) with the AI-detection findings, calls them 'convergent,' and treats the combination as sufficient. This is institutional reasoning of a kind designed to produce a clean outcome without requiring proof. Le Éditions du Boréal, the Canadian publisher, explicitly declined to reach that conclusion: 'Without credible, concrete evidence, it is not our place to decide whether this book has a right to exist.' That is a more honest epistemological position than the one the Goncourt adopted. Fitoussi's stated motivation deserves scrutiny. He told his own X followers he launched the campaign not because he was opposed to AI but because he feared AI would 'cause the disappearance of reading and writing' — a concern that led him, using an anonymous account, to target a Black Caribbean novelist on the day that novelist won a major prize. Orélien has characterized the campaign as racially motivated. Whether or not that characterization is legally or morally precise, the pattern — anonymous account, unreliable detection tool, right-wing columnist, prestigious prize body flinching — is legible enough to name. The prize mechanism itself is the lasting story. The Goncourt shortlist publishes October 6; the winner is announced November 4. What the Academy has demonstrated is that an anonymous social media account, wielding a contested detection tool, can remove a book from consideration at the highest level of French literary culture within weeks, without the author being given meaningful opportunity to contest the evidence. That is not a procedure. It is a panic. The plagiarism allegation may ultimately prove substantive — that is a separate and legitimate question. But bundling it with junk-science AI detection to produce a single withdrawal statement is not how institutions that claim to 'preserve integrity' are supposed to operate.
