Sounds familiar
I don't like the unnecessary defensiveness by the author to pangram proof.
Excellent use of AI? Disclose it and avoid the fuss.
If I had a book that was hoping to win this award, anonymous accusations like this seem like a pretty easy way to shift the odds. It's impossible for the author to disprove the allegations and the award committee will want to 'avoid controversy'
Anyway, here's the thread from twitter if people are interested: https://x.com/Pangramed/status/2102383484027171306?s=20
[^1]: https://www.independent.co.uk/arts-entertainment/books/featu...
[^2]: https://www.thetimes.com/uk/education/article/jason-arday-re...
Critics say the author almost exclusively relied on AI to create his work, based on Pangram’s findings.
Pangram says it “is trained specifically to detect the latest AI models from frontier labs like OpenAI, Anthropic, Google, Meta, DeepSeek, xAI, and more. [They] benchmark against 26 models (so far)…”.
I can probably list more than 26 models off the top of my head. Granted, some models have been distilled against others and will be “detectable” but that’s NOWHERE near a large enough sample size to be reliable in 2026.
You or I could get a flag by that detector in our first attempt, and it wouldn’t take long for a false positive against enough attempts of truly original thought either.
The only thing I’d actually believe is things like Claude’s text watermarking system. However, those systems will never be present for EVERY model, so reliable blanket detection remains an impossibility.
The entropy in written text of meaningful length is simply too great for detection to be reliably possible in 2026, especially considering the changes the author can make.
"Experimenting, I’ve also found that I can pretty reliably get Pangram to declare large sections of text to be AI-written or human-written depending on the larger textual context I place those sections in - that is, I can take AI-generated text and embed it in human-generated text and have Pangram declare it 100% human, and I can take human-generated text and embed it in AI-generated text and have Pangram declare it 100% AI. I’ve also found that it’s pretty easy to induce false positives, that is, to write texts myself that Pangram identifies as AI-written. All of this makes me hesitant about the Pangram tool, and in particular the way many people use it, as a one-shot gotcha machine."
Sounds like it may be very easy for someone motivated enough to find sections, even longer sections, that turn out as 100% AI written.
[1] https://www.lefigaro.fr/culture/qui-se-cache-derriere-le-com...
https://www.lapresse.ca/arts/litterature/2026-09-23/thelyson...
* Media ran the entire novel through Pangram and got a 94% AI score [1], whereas all other Goncourt finalists, Haitian/African novels and French classics they tested came out as 100% human. I recall they also tested works specifically from before and after the AI boom and found a shift, but I don't remember where I saw that.
* People on social media, as well as traditional media, have then found cases of blatant plagiarism in Orelien's work a decade ago (this includes one prized short story and several posts to blogs and media) [2]
I think we have to be careful not to assume that tools like Pangram are completely accurate and the word of God, but we can't dismiss them outright, and this particular author's history of plagiarism makes it much harder to give him the benefit of doubt.
[1] (French) https://ici.radio-canada.ca/rci/fr/nouvelle/2286476/thelyson...
[2] (French) https://www.lapresse.ca/arts/litterature/2026-09-23/thelyson...
Comparing to other Haitian authors was also relevant, because another defence from the author was that his "caribbean style" could have tripped up the detector.
But yes, other than that, that's a good point. My anecdotal evidence is that my own unpublished writing, both in French and English, contains no false positives, but even in that case, that's only evidence that my style isn't AI-adjacent. I'd be curious to see concrete examples of false positives, but it's hard to fully trust these if they don't have proven timestamps.
I.e. they're using AI output a black-box truth oracle, without understanding how it works, or being able to independently reproduce its conclusions as human thinkers.
Perfectly healthy habit for society to normalize.
Lots of short sentences. Lots of weird metaphors. Many infamous "it's not x, it's y".
"Why do you care is there is horse meat in your beef burger patties if it tastes good" vibe
Would we care if someone used a typewriter vs a computer to write it.
This looks like a weird snobbishness. The fact that it’s by Europeans makes it more funny since they have been left out of the AI race.
In the end you have a final product. Can critique it on its own. He used ai to help write it and you couldn’t tell before, says more about the product and perhaps you don’t like what it says about you .
Then again. Europeans have ret-conned genericized names into terroir so it's not exactly surprising.
[1] https://www.lapresse.ca/arts/litterature/2026-09-23/thelyson...
I don’t think anyone is saying skin color conveys immunity from accusation. But I hope you can see that the odds that a dramatically higher per capita rate of accusations against people with a particular skin color should at least give pause.
Why would they target these specific 3 black authors? I assume there are hundreds/thousands of successful black authors, 3 "this year" seems very small to call it a targeted attack
Race has not been a part of the public conversation. Not sure if it would have been that way if the guilty person was non-white (but I'm slightly optimistic, given that the first person found guilty was a woman and gender was not part of the discussion).
By whom?
A simpler (and easy to think of) explanation is normal systemic racism. Black writers are likely underrepresented in the training data begin with, so the models assume Black writing can't be real.
Pangram is unreliable on short text, but much more reliable on long form text.
unified101•1h ago
rcxdude•1h ago
unified101•48m ago
rcxdude•19m ago
fidotron•56m ago
It's entirely plausible detection is totally different in different languages.