The Goncourt and the machine

On 3 November the Académie Goncourt gave its prize to Philippe Jaenada for L’Inconnue du quai de Javel, by six votes to three. The next day he told France Inter that he had used ChatGPT three times while writing it. One sentence in the book was written by the machine, and he put it there on purpose.

The admission came less than six weeks after the same academy had removed another novel from its selection, saying it was very likely largely the product of artificial intelligence. The two cases look alike. We think they sit on opposite sides of a line that runs not through the tool but through the person using it.

What he said

Jaenada, 62, won for his investigation into the unsolved murder of a young Breton woman in Paris in the late 1940s. The first of the three uses was the sentence, which he inserted, he said, to “jouer avec le lecteur”, to play with the reader. He has revealed which sentence it is only to the Goncourt jurors, and joked: “Si ça sort, je sais de qui ça viendra”, if it gets out, I’ll know who it came from.

The second time, he said, was “pour me moquer un peu de ChatGPT, qui fait des réponses complètement absurdes à propos d’une donnée technique”: to make a little fun of a tool that gave absurd answers about a technical detail. The third was research. “On ne peut pas faire comme si ça n’existait pas,” he said, we cannot act as if it did not exist. And: “Il faut s’en servir, il faut le reconnaître, il faut l’apprivoiser et ne pas se faire remplacer.” In our translation: we have to use it, admit it, tame it and not let ourselves be replaced.

Reactions online were mixed, Euronews reported. Some saw AI becoming as unavoidable as a spellchecker, others were disappointed.

Two books, one academy

On 25 September the academy had withdrawn C’était ça ou mourir, the first novel by Thélyson Orélien, from its first selection. It cited plagiarism by the author and analyses by researchers and journalists concluding that the book was “selon toute vraisemblance très largement le produit d’une Intelligence Artificielle”, in all likelihood very largely the product of AI. Its role, the academy said, is to promote and crown “la littérature écrite par des femmes et des hommes”, literature written by women and men.

Orélien denied using AI and told AFP that “aucune phrase de ce livre n’a été créée par une machine”, not one sentence of the book was made by a machine. On 22 October his Canadian publisher, Les Éditions du Boréal, said that artificial intelligence had been used in writing the novel, and ended its work with him.

Side by side, the two cases differ less in the amount of machine text than in what happened around it. In one, a book was judged largely generated while its author denied it. In the other, a writer chose one sentence, told the jury which one, and explained himself on the radio the day after he won.

Where the line sits

Nothing in Jaenada’s account suggests the tool decided anything. He chose to ask for a sentence, chose to keep it, chose where it goes and why. Mocking an absurd answer turns the tool’s failure into material. Using it for research is what writers already do with archives and search engines, and his own example shows why every answer needs checking.

That is where we would draw the line between a tool and an author. It does not run through the share of words a machine produced. It runs through two questions: did a person choose every sentence that stayed, and will that person answer for it? A writer who asks a model for ten lines and keeps one is still the author. A book in which the model chose and the person only signed is something else, whatever the percentage.

Hiding is the problem

If AI in creative work has a scandal, it lies in concealment, not in use. A reader, a jury or a client trusts that the name on the work belongs to the person who made the choices. That trust survives a tool. It does not survive finding out later that the tool did the work and someone kept quiet.

We read the academy’s decision in September and Jaenada’s openness in November as two sides of one principle: the name on a book should tell the truth about who made it. Disclosure does not have to be solemn. Jaenada made his part of a game with the reader, and told the jurors where the sentence was. He used the tool, kept the choice and said so, in that order.

The same question for brands

A brand that publishes images, film or copy made with AI makes the same promise as an author: this is what we chose to say. The EU now writes part of that into law. Since 2 August 2026, Article 50 of the AI Act requires anyone who uses AI to make a deepfake to disclose that the content is generated or manipulated, and requires the same for AI generated text published to inform the public on matters of public interest.

The duty for text falls away where the content has gone through human review or editorial control and a person or company holds editorial responsibility for it. The law draws its line where we would: at someone who reviews what goes out and answers for it.

The other half of the question is what went in. Last week the owners of Elvis Presley’s music and Eminem’s publisher, Eight Mile Style, sued the music generator Suno in California, accusing it of copying their works and, in the Presley case, of imitating his voice. Suno said its service is there to help people make original music, “not to trade on anyone’s name or likeness.” For a company, both halves come down to a record: which images, clips and texts were made with AI, the references and versions behind them, and the person who approved each one.

Where crue stands

In Reel, which is live, the ideas, references and styles to lock are agreed with a person before anything is made, a person picks every image and clip that is used, and each one sits in a media library with every version and the recipe behind it. In Studio, in early access, the assistant drafts and a person approves every change before it goes live, with every action in an audit log. Neither makes the choice for you. Both make it possible to say, at any time, who chose.