When a novel titled C’était ça ou mourir hit the shelves in late August, it quickly became one of the season’s literary highlights. The book, published by Grasset, tells the story of a Haitian teacher who escapes gang violence, journeys to Canada, and finally reclaims the simple right to breathe without fear. Its vivid rhythm, Caribbean imagery, and raw emotional cadence earned it praise from critics, led to selections for several autumn literary prizes and, on a Monday, secured the coveted Prix du Roman Fnac.
Detection claim and the rise of Balance ton Claude
Only days after the award ceremony, an anonymous X account created in June under the name Balance ton Claude—a reference to Anthropic’s AI assistant Claude—released a series of excerpts from the novel to a detection service called Pangram. The tool returned a striking verdict: “100% IA.” The same account fed older passages from the author’s previous work to the same system, which this time flagged them as human-written. The contrasting results spread rapidly across social media, prompting the writer to issue a denial and the publisher Grasset to describe the allegations as “atrociously injurious.”
How AI detectors work and why they are contested
Modern detection software no longer compares texts against a database of known AI output. Instead, it analyses statistical patterns—repetitions, phrasing, syntax—that are typical of language models. When a passage matches those patterns, the system assigns a probability that it originated from an artificial source. Critics note that such probability scores can be misleading: a high figure does not prove machine authorship, merely that the text resembles the model’s training data. Independent studies have shown false-positive rates that vary widely between tools, and the same sample can be judged differently by two detectors. Moreover, many programs are calibrated on standard French or English, struggling with regional dialects, Caribbean Creole influences, or the idiosyncratic style of a writer like Orélien.
Historical echoes of authorship doubt
The current dispute echoes earlier moments when the authenticity of a voice was questioned. In 2014, journalists travelled to Hallencourt in the Somme to verify whether Édouard Louis’ gritty account of rural poverty matched the lived reality of its inhabitants. More than a decade later, the same scrutiny is carried out from a chair, using a click-and-paste routine on Pangram. Another precedent is the 1975 revelation that the celebrated author Émile Ajar was, in fact, a literary pseudonym for Romain Gary, who used a stand-in to circumvent the Goncourt prize rules. Those cases focused on the *identity* of the writer; the present case adds a new layer—whether the *hand* that typed the words belongs to a human or a machine.
Implications for literary creation and future disputes
If the detector’s judgment were accepted, it would mean that a work already lauded by juries and readers could be dismissed as a product of an algorithm. Yet the reception of the novel—its emotional impact, its thematic resonance—remains unchanged regardless of the tool used to draft it. The broader debate now centers on what degree of machine assistance is permissible before the label “author” loses meaning. Does asking an AI for synonyms count, or does commissioning an entire manuscript via a prompt cross a line? The industry has long embraced digital assistants for research, translation, and editing; the controversy forces publishers, award committees, and readers to reconsider the contract that defines originality.
For Orélien, born in 1988 in Gonaïves, Haiti, and later based in Québec, the stakes are personal as well as professional. He has assembled a dossier with his editor to demonstrate the manuscript’s chronology, citing a first draft completed in —well before large-scale text-generation models were publicly available. The case continues to evolve, with the novel slated for translation into around twenty languages and its story already resonating across continents. Whether the detection claim will hold or fade, the episode has already highlighted the fragile trust placed in algorithms that claim to police creativity.
What the literary world may watch next
As AI tools become ordinary companions in the writing process, the sector is likely to develop clearer guidelines on disclosure and attribution. Until then, each high-profile accusation—whether involving a debut novelist from Québec, a celebrated French author, or a short-story winner from the Commonwealth—will serve as a test case for how far the industry is willing to let machines shape its most intimate expression.
