A celebrated novel about migration, a prestigious French literary prize, and an AI detection tool that returned a 100% probability of computer-generated text. The case of Thélyson Orélien, the Haitian Canadian writer whose publisher has now severed ties with him, is not simply a story about one author and one book. It is a stress test for how the literary world, and the institutions that govern it, handle a problem that has no clean resolution.
What the Metadata Revealed
The sequence of events matters here. Orélien’s novel, C’était ça ou mourir (It Was Either That or Die), was withdrawn from the Prix Goncourt list after AI detector Pangram indicated a 100% probability that the text contained computer-generated content. His Canadian publisher, Les Éditions du Boréal, initially stood by the 38-year-old writer. That position changed within days.
Boréal conducted a further investigation, gathering all working versions of the manuscript and hiring an outside party to analyse the drafts. The results were specific: traces of AI tags were found in the metadata of certain documents. This is a technical detail worth pausing on. Metadata, the hidden layer of information embedded in digital files, can record which software was used to create or edit a document. It does not record intent, and it does not record how much of the final text was shaped by a machine. Boréal acknowledged this directly, stating that it did not know the extent of AI’s use in the writing process and that a definitive answer would likely never be available. The publisher ended its professional relationship with Orélien on the basis of confirmed traces, not confirmed scope.
Orélien, for his part, has not conceded the central allegation. He told the French newspaper Libération that he used AI “in the creative process” for “historical research” and, separately, acknowledged using tools for foreign idioms. He maintains that no sentence in the book was created by a machine. He also noted that he began writing the novel in late 2017 and completed a first draft of more than 300 pages in 2019, several years before services such as ChatGPT became widely available. That timeline is not irrelevant. It complicates any assumption that the book was assembled with modern generative AI from the start.
The Gap Between Detection and Proof
This is what most coverage of the case tends to flatten: there is a significant distance between “AI was used” and “AI wrote this.” Boréal’s own statement captures the problem honestly. The publisher confirmed that AI was used in the writing process but could not determine how extensively. A 100% score from an AI detector is not the same as a forensic audit. Detection tools, including Pangram, are probabilistic instruments. They identify patterns associated with machine-generated text. They cannot reconstruct a creative process or assign authorship percentages.
The novel had been scheduled for translation in more than 20 countries. That scale of international interest signals that readers and publishers found genuine literary value in the work. None of that context resolves the ethical question, but it does illustrate the stakes. The question is not only whether AI was used. It is what kind of use crosses a line, and who gets to draw that line.
Orélien’s situation is further complicated by a separate, earlier allegation of plagiarism, to which he admitted making “unacceptable mistakes” in his past writing, without specifying what those mistakes were. He attributed them to youth and impatience. That admission, while unrelated to the AI question, has shaped how the current allegations are being received.
Why This Case Will Not Be the Last
The Orélien affair surfaces a structural tension that literary institutions are not yet equipped to resolve. The Prix Goncourt, one of France’s most prestigious book prizes, withdrew the novel from consideration. A publisher conducted a technical investigation and ended a professional relationship. An author denied the core allegation while acknowledging partial use of AI tools. None of these actions produced a definitive account of what actually happened inside the writing process.
This is the deeper issue. As AI tools become more embedded in creative workflows, the boundary between assistance and authorship becomes harder to locate. Using AI to research historical facts or translate idioms occupies a different ethical register than using it to generate prose. But detection tools cannot distinguish between these uses. Publishers, prize committees, and readers are being asked to make judgments that the available technology cannot fully support.
What the Orélien case makes visible is the absence of shared standards. There is no agreed framework for how much AI involvement disqualifies a work from literary consideration, no established protocol for what evidence a publisher should gather, and no consensus on what an author owes readers in terms of disclosure. The institutions involved here improvised their responses in real time, and the results were inconclusive by their own admission.
In Short
The Orélien case is not a verdict. It is a demonstration of how unprepared literary institutions are for a question that will keep returning. AI detection tools can flag patterns; they cannot reconstruct intent or measure creative contribution. Publishers can investigate metadata; they cannot determine scope. What the case leaves behind is not a clear answer about one novel, but a clear signal that the absence of shared standards around AI and authorship is now a practical problem, not a theoretical one.
Based on reporting from The Guardian.