French literary circles were jolted this month when the Goncourt Academy removed Thélyson Orélien‘s widely sold novel from its shortlist after an anonymous social-media account said the book was largely generated by AI. The novel, published in French as C’était ça ou mourir and translated as It Was Either That or Die, has topped sales charts and reportedly sold about 35,000 copies; it also received the Fnac novel prize days earlier. The allegation rested on analyses run with the Pangram detection tool and prompted the academy to cite consistent findings from researchers and journalists as the basis for withdrawal.
Thélyson Orélien, a writer who left Haiti for Canada after the 2010 earthquake, rejects the accusation and says he drafted the book between 2017 and 2019. His publisher backs that timeline. In parallel to the AI claims, Quebec newspaper La Presse revisited several of Orélien’s earlier columns and reported instances of overlapping passages with other authors; a Radio-Canada review identified at least five pieces with partial copying. Orélien has said he is compiling documentation with his publishers to contest the charges and has described some responses as racist attacks.
Pangram, the detector cited in the case, is regarded by some studies as highly accurate, with published work reporting detection rates above 95% and few false positives. Nevertheless, experts note limitations and contextual pitfalls. French analyst Benoît Raphaël tested Pangram on several Haitian novels and on Orélien’s pre-2023 texts and found them identified as human-written, while Raphaël and others also flagged stylistic markers they say can resemble AI output. Separately, columnist Samuel Fitoussi has acknowledged involvement in a group behind the social-media account that raised the alarm, complicating the public narrative.
Voices across the literary field have reacted strongly. Author Virginie Despentes publicly defended an author’s freedom to use tools without scrutiny, while Raphaël Doan and scholar Alexandre Gefen framed the episode within evolving notions of authorship and technique. The case has intensified debate over the reliability of automated detection, the ethics of undisclosed assistance, and readers’ expectations of a human creative voice. For now, the episode leaves awards bodies, publishers and readers navigating unsettled ground over how to assess and disclose the role of new writing technologies.