Pangram flags 99% of an award-winning Peruvian novel as AI, but publisher stands by author

Pangram flags 99% of an award-winning Peruvian novel as AI, but publisher stands by author

The AI detection tool Pangram flags 99 percent of the award-winning novel "Donde el agua trabaja" by Peruvian author David Silva Rojas as AI-generated. That is a detector result, not a confirmed finding that AI wrote the book.

Despite the allegations, the Mexican publisher Almadía is standing by the award after reviewing documents from the author. This comes from an essay in The Point Magazine. The newspaper El País adds that Pangram had also attributed more than 80 percent of the text from interviews, an essay and the author's Instagram denial to AI.

The essay's author makes several arguments. Latin American critics, the essay says, ignored Pangram's test results and dismissed the allegations wholesale as racist. The essay compares the debate to the confirmed use of AI by Thélyson Orélien, and to accusations involving the Caine Prize and Jamir Nazir, who kept his award. It also argues that literary juries often favor clichéd stories about minorities, which language models can easily imitate.

The wider industry is already reacting. Publisher Gallimard reportedly plans to screen manuscripts with Pangram starting next year. In the US, the Authors Guild, which represents writers' interests, warns against using AI detectors as the sole basis for decisions, even though Pangram reliably identified human-written texts as such in its tests.

Key facts

  • Pangram flags 99 percent of "Donde el agua trabaja" by Peruvian author David Silva Rojas as AI-generated; this is detector output, not confirmed AI use.
  • Mexican publisher Almadía stands by the award after reviewing documents from the author, per an essay in The Point Magazine.
  • El País reports that Pangram also attributed more than 80 percent of text from interviews, an essay and the author's Instagram denial to AI.
  • Gallimard reportedly plans to screen manuscripts with Pangram starting next year.
  • The Authors Guild warns against using AI detectors as the sole basis for decisions, even though Pangram reliably identified human-written texts as such in its tests.

Why it matters

A detector's verdict has landed on a prize-winning book, and the publisher has chosen to side with the author instead. The case shows how an AI-detection score can become a public accusation, and how the dispute then spreads into arguments about racism, literary juries and how reliable the tool itself is. It also lands as Gallimard reportedly plans to screen manuscripts with Pangram starting next year, so detectors may soon be part of publishing routine.

Who it affects

Authors, whose work can be publicly questioned on the strength of a detector score. Publishers and prize juries, who must decide how much weight to give such scores. Critics, especially in Latin America, whom the essay accuses of dismissing the results as racist. And writers' organizations such as the US Authors Guild, which cautions against relying on detectors alone.

How to use it

The source offers no usage guide for Pangram. The one practical lesson it reports comes from the Authors Guild: do not use AI detectors as the sole basis for decisions. Almadía's approach, as described, was to review documents from the author before keeping the award.

How solid is it

The core facts are secondhand. The Decoder relays an essay in The Point Magazine and a report from El País; the 80 percent figure for interviews, an essay and the Instagram denial is El País's claim, and the essay's arguments are the essay author's own views. The source does not say who ran the Pangram test on the novel or when. The source gives no confirmation that the novel was in fact written with AI; 99 percent is a detector result. The source does not say what documents Almadía reviewed.

Risks and caveats

A 99 percent flag is not proof of AI authorship, and the source does not confirm it. Detector scores on interviews and an Instagram post show how the same tool can be turned on an author's other writing. The Authors Guild's warning applies even though Pangram reliably identified human-written texts as such in its tests. The comparisons in the essay to Thélyson Orélien and to the Caine Prize and Jamir Nazir cases are the essay author's framing. The source does not say what the Caine Prize and Jamir Nazir accusations concerned or what Orélien's confirmed AI use was. Gallimard's plan is reported with "reportedly" and is not confirmed in the source.