Pangram founder claims AI detection in bestselling novel; skeptics question reliability
Max Spéro says his AI detection tool flagged Thélyson Orélien's novel, but researchers challenge Pangram's accuracy against adversarial techniques.
What to know
- Pangram founder Max Spéro claims his AI detection tool identified AI use in a bestselling novel by Thélyson Orélien, positioning the tool as a potential standard for literary authenticity.
- Critics cite published research showing Pangram can be circumvented by requesting specific writing styles from AI systems or using humanizer tools, and argue no detector can be 100% reliable.
- Debate centers on the gap between detection and proof: commenters distinguish between Pangram flagging text and establishing intentional AI authorship, and note AI use as a drafting aid differs from sole reliance on AI.
The dispute Whether Pangram's detection constitutes credible evidence of AI authorship or merely a preliminary signal that requires corroboration against potential false positives and adversarial evasion techniques. · positions read across 7 posts and comments
Pangram is a reliable detection tool superior to subjective judgment and warrants serious consideration in authorship disputes.
-
“Pangram is very, VERY good at what it does. I'd much rather have AI accusations based on evidence like that than based on feelings.”
TinyTricksterTanuki · reddit ↗
Pangram can be fooled by adversarial techniques and research shows high unreliability; detection alone doesn't prove intentional AI authorship.
-
“There are now quite a bit of third-party research showing that it is quite easy to fool Pangram into believing AI-written texts are human-written by asking the AI to write in particular styles or feeding the AI output through an «…”
NekoCatSidhe · reddit ↗
“it could come to be 'like a nutrition label' for written prose”
Max Spéro, Pangram founder · Le Monde EN ↗ · Sep 25, 1:00 PM
Max Spéro Pangram founderThélyson Orélien Novelist
How it unfolded 5 developments, newest first · click a bar or a number to jump articlespostscomments
-
5
Researchers cite published work showing Pangram can be circumvented
A commenter points to third-party research demonstrating that Pangram can be fooled by requesting specific writing styles from AI or running output through humanizer tools, and notes high false-positive rates and insufficient research on the reverse problem.
“There are now quite a bit of third-party research showing that it is quite easy to fool Pangram into believing AI-written texts are human-written by asking the AI to write in particular styles or feeding the AI output through an « humanizer AI ».”
— NekoCatSidhe -
Actually no. There are now quite a bit of third-party research showing that it is quite easy to fool Pangram into believing AI-written texts are human-written by asking the AI to write in particular styles or feeding the AI output through an « humanizer AI ». So it is only going to catch the dumbest of AI-using authors in the first place (not to…
1 more of the top 2 · 2 posts in this stretch
-
ngl pangram founder got no doubt cause its his own detector show actual passages from the book and let someone else check that shit before we treat a high score like a confession
-
-
4
Critics distinguish between detection and proof of intentional AI use
Commenters argue that a tool's "no doubt" is not the same as definitive evidence, and note that AI can be a legitimate drafting aid rather than sole authorship.
“No doubt isn't the same as evidence. I only start doubting an author when I see the same cadence in three separate chapters, not when one publisher says it.”
— ShortVermicelli1719 -
No doubt isn’t the same as evidence. I only start doubting an author when I see the same cadence in three separate chapters, not when one publisher says it. AI can be a drafting tool, but that’s a different claim.
-
-
3
Supporters argue Pangram evidence is stronger than subjective judgment
A commenter defends Pangram as superior to intuitive assessment, arguing multiple flags on long-form work by the same author constitute "pretty overwhelming" evidence, though warns against single-snippet accusations.
“I'd much rather have AI accusations based on evidence like that than based on feelings about how something is written.”
— TinyTricksterTanuki -
Why not? Pangram is very, VERY good at what it does. I'd much rather have AI accusations based on evidence like that than based on feelings about how something is written. Obviously, no one should join a witch hunt based on one single Pangram report of a short snippet or anything. But if Pangram flags multiple long form pieces by the same writer…
1 more of the top 2 · 2 posts in this stretch
-
I see these claims constantly. Can you share any evidence of Pangram flagging a text like those mentioned as AI?
-
-
2
Users debate Pangram's technical capabilities and limitations
Reddit commenters contest the reliability of Pangram's detection. Some defend its effectiveness for long-form analysis, while others argue research shows it can be fooled through specific writing styles or AI humanizer tools, and note the lack of peer review on adversarial testing.
“Pangram doesn't rely on LLMs though. These are entirely different kinds of AI.”
— Zonoro14 -
Pangram doesn't rely on LLMs though. These are entirely different kinds of AI. The task of "write indistinguishably from a human" and "detect whether a given text was written by an A unable to perfectly imitate human writing" have very little in common, and the latter task is vastly easier.
1 more of the top 2 · 2 posts in this stretch
-
Suddenly readers are in favor of AI. What a day to be alive. Not a good day, but a day.
-
-
1
Max Spéro claims Pangram detected AI use in Orélien's novel
Le Monde reports that Pangram founder Max Spéro has stated he has "no doubt" about AI use in Thélyson Orélien's bestselling novel. Spéro describes Pangram as potentially becoming "like a nutrition label" for written prose.
-
2 outlets first by Le Monde EN, 1d ago · also Le Monde.fr · read ↗
-