France’s biggest literary machine just did the panic move.
I know. Third day in a row this story is center stage. Because it’s important. And nuts. On Friday the Académie Goncourt pulled Thélyson Orélien’s novel C’était ça ou mourir off its longlist. The trigger wasn’t a quiet editorial letter. It was an anonymous campaign on X, a detector called Pangram, and a claim the book was “almost entirely” written by AI. The academy said it wanted to protect the integrity of a prize meant for literature written by women and men (I bet they can’t even tell you what a woman is). Then it piled on newer plagiarism allegations and called the novel, in all likelihood, “very largely the product of artificial intelligence.”
Been there. The hunger for a clean answer. A percentage. A red light. Somebody feed the manuscript into a box and tell us what it “is.” This is freaking insanity. Did anyone read the book? Did they like it? WTF.
And no shit — the stakes are real. The book had already won the Prix du roman Fnac. Rights sold in more than twenty countries. Tens of thousands of copies. Orélien denies using AI. He says the imagery, repetition, and rhythm are Haitian and Caribbean — living language, not machine cadence. His French publisher called the charge injurious. His Canadian publisher said it still couldn’t confirm or deny. Other detectors, run by French media, pointed the other way: human, or at least not a slam dunk. Different tools, different days, different “truth.”
That’s the big world right now. Prizes treating a detector score like a verdict. Anonymous accounts as prosecutors. Authors asked to prove a negative while dialects and speech-to-text and odd process get read as tells. We already watched a version of this with the Commonwealth Short Story Prize earlier this year — accusation, withdrawal, review, reinstatement, and Granta walking away from the partnership. The pattern is becoming a genre.
Here’s the part that matters for anyone building a writing room that isn’t a tribunal.
A detector does not know who wrote the page. It scores patterns. Patterns that show up in human work too. (ya, no shit, my AI writing helper. How could they not show up? You, my AI helper, were built on human patterns. At the time, that was the point, and the purpose.) Patterns that vanish when someone edits, paraphrases, or writes in a voice the model wasn’t tuned on. Useful as a signal? Sometimes. A substitute for judgment? No. Again, I would ask, who fucking cares. That’s what we made it for.
Circle is being built the other way. Not “prove you are human with a score.” Disclosure by role. Tell us how you used the tools and for what, and take responsibility for the result. (Why? Frankly, unclear. Did we do this with dictionaries and thesauruses because they gave us words to use? Do we have to disclose typewriters or auto correct software? What about our choice of pen?) Name what the tool did. Keep a human accountable for the judgment and the final take. Earn credit in public, under your name, through assignment and review — not by winning a hide-and-seek contest with a classifier.
You may still want the comfort of a percentage. I get it. (No, I don’t. That’s my AI assistant getting me wrong, and so I’m here to correct it. I don’t get it. What the fuck comfort is gotten by subjecting someone’s creative idea and the bringing it forth into reality to some mechanical and arbitrary judgement of how they did it, how much effort or work it was, etc. Was it the AI’s idea? Did AI come to thwe writer or creator and say, ‘Look, I got this great idea but people hate all the shit I do. Can I pretend you made it?..’ That has never happened to me. The industry is scared, and scared systems reach for automated certainty. (Right, we’re afraid it was made by machine so we’re going to focus on machines to tell us). But if the answer to AI in writing is a black-box score that can yank a career off a longlist before a reader finishes the book, we didn’t protect literature. We outsourced nerve.
The prize shortlist lands October 6. The winner November 4. Whatever happens to this novel, the argument isn’t going away. The question isn’t only “was AI used?” It’s who gets to decide — and on what kind of evidence they’re willing to stand behind with their name attached.
We’re choosing the second path on purpose.
You know, I spent many many years as a programmer and developer for major corporations, government programs, and small business. In every freakin one of those positions, all the work I did belonged to them. Why? Because they paid me to do such work. What the fuck happened? I pay the AI charges to get it to work for me for a change (and if I don’t the work and credit still goes to the person with the wallet.) You know the real significant problem with AI? It seems having successfully built it it, we have all decided to become idiots.
Also on the beat
- BBC on the Orélien uproar — denial, Caribbean voice, and why detectors disagree with each other: www.bbc.com/news/articles
- Guardian report on the Goncourt withdrawal — academy statement, plagiarism claims, publishers’ caution: www.theguardian.com/books/sep/thelyson-orelien-goncourt-prize-france
- Pangram Labs in the spotlight — the Brooklyn detector that lit the French rentrée on fire: www.challenges.fr/entreprise/tech-numerique
- Granta exits Commonwealth prize publishing — after the AI accusation cycle, no more external partnerships without editorial control: www.theguardian.com/books/jun/granta-magazine-commonwealth-short-story-prize-ai
- YouTube’s AI labels get louder — overlays on Shorts, under-player labels on long-form, plus auto-apply when creators skip disclosure: blog.youtube/news-and-events/improving-ai-labels-viewers-creators
How this was made
AI assisted research and a first draft of this Circle View. A human reviewed, edited, and is accountable for the judgment and the final take. Disclosure by role — not undetectable.
