Canadian publisher Boreal has severed its relationship with Haitian-Canadian writer Thelyson Orelien after an investigation confirmed artificial intelligence was used in the writing of his novel "C'était ça ou mourir" ("It Was Either That or Die"). The publisher gathered all working versions of the manuscript and hired an outside party to analyse drafts, finding "traces of AI tags in the metadata of certain documents." Boreal acknowledged it does not know the extent of AI's use and "will likely never have a definitive answer." Orelien, 38, had been on a meteoric trajectory. The novel about migration was scheduled for translation in more than 20 countries. France's top literary prize, the Goncourt, had placed it on its shortlist — until the AI detector Pangram returned a 100 percent probability score for computer-generated text. The Goncourt pulled the book. Boreal initially stood by Orelien as recently as last week before reversing course. Orelien maintains a specific defense: he used AI tools "in the creative process" for "historical research" and "foreign idioms," but insists "not a single sentence in this book was created by a machine, nor the meaning or the structure of a story that comes from my gut." He told Libération he began writing in late 2017, before ChatGPT existed. He called the allegations "slander and heresy" and suggested racism was a factor. The evidentiary basis is worth scrutinising. Boreal's case rests on two things: metadata AI tags found in document files, and Orelien's refusal to provide further information after initial cooperation. Pangram's 100% score drove the Goncourt's decision. But AI detection tools remain probabilistically contested — a 100% confidence score from a commercial detector is not the same as a forensic finding. The metadata tags are more concrete, but Boreal itself admits it cannot determine the extent of AI involvement. What is emerging is a new institutional regime in publishing: AI detection as gatekeeper. Pangram, a commercial product, effectively held veto power over a Goncourt shortlist nomination and a 20-country translation deal. The speed of institutional reaction — shortlist removal, publisher severance — suggests that the mere signal of AI use now carries reputational destruction regardless of nuance. Orelien's defense that he used AI for research but not composition is a distinction the current detection infrastructure cannot make. The racial dimension Orelien raised deserves honest assessment rather than dismissal. A Haitian-Canadian writer producing a breakthrough literary novel about migration is statistically unusual in French-language publishing. Whether or not racism motivated the initial allegations, the outcome is that a writer from a marginalized background has been stripped of a career-defining achievement based on tools whose accuracy and granularity remain hotly debated in the technical community. The 20-year trajectory here is stark: if commercial AI detectors become the de facto authenticity arbiters for literary publishing, the power to validate or destroy a writer's career shifts from editors and critics to opaque algorithmic tools. Writers who use AI for any part of their process — research, translation, ideation — face binary guilty/not-guilty verdicts from systems that cannot distinguish between writing with AI and writing by AI.
