France’s literary establishment has been drawn into controversy after Haitian-Canadian author Thélyson Orélien, whose debut novel has won major recognition, was accused of using artificial intelligence to write the work.
Orélien has denied the allegation, while his publishers in France and Canada have publicly backed him as questions over AI authorship and detection technology intensify.
The dispute began after the little-known X account Balance ton Claude said passages from C'était ça ou mourir, or It Was Either That or Die, had been analyzed using the AI detection program Pangram.
The group claimed the novel was written almost entirely by artificial intelligence and said tests of several passages produced results indicating AI authorship.
Orélien rejected the accusation and told French newspaper Libération that he was preparing a dossier with his publishers in France and Quebec to demonstrate that he wrote the novel.
He said his first draft dated to 2019, before text-generating AI programs became generally available.
French publisher Grasset condemned the allegations and reaffirmed its confidence in Orélien, while Canadian publisher Éditions du Boréal said it stood alongside the author.
Orélien, who was born in Haiti and moved to Montreal in 2010, has said his writing reflects Haitian and Caribbean traditions, including their imagery, repetition and rhythms.
The novel follows a Haitian migrant fleeing violence and traveling through Latin America and the United States before reaching Canada.
It has received several awards and recognition, including the Prix du Roman Fnac, while also being considered for major French-language literary prizes including the Goncourt.
Philippe Claudel, president of the Académie Goncourt, said no decision had been made on whether to remove Orélien’s novel from consideration.
He said the academy could examine the matter if indisputable information emerged before its October 6 meeting.
The controversy has also renewed scrutiny of AI detection systems.
Critics say such programs can incorrectly identify human writing as AI-generated or fail to detect AI-produced material, particularly after editing or paraphrasing.
Pangram says its system is 99.7% accurate, while Balance ton Claude has defended its reliance on the technology.