Oossa

Pangram flags most of award‑winning Peruvian novel as AI‑generated

AI detector Pangram says 99% of David Silva Rojas’ novel looks machine‑written, but publisher Almadía backs the author after reviewing his documents.

By Published by Oossa: 2 min read

Clark Young · Unsplash

The AI‑detection service Pangram has reported that 99 percent of the text in "Donde el agua trabaja," the novel that won Peru’s major literary prize, looks like it was generated by an AI model. The claim sparked a heated debate in the literary community. The publisher, Mexico‑based Almadía, says it has examined the author’s original drafts, correspondence and other paperwork and remains convinced the book is the work of David Silva Rojas. The publisher’s stance was detailed in an essay published by The Point Magazine.

Pangram’s analysis did not stop at the novel. The tool also marked more than 80 percent of the author’s interview transcripts, a personal essay and even his Instagram denial of the allegations as AI‑generated, according to the Spanish newspaper El País. Critics of the detector argue that such high false‑positive rates can damage writers’ reputations, especially those from minority backgrounds. The author himself wrote an essay accusing Latin‑American critics of dismissing the detector’s results as racist and of favoring clichéd stories that language models can easily mimic.

How the publisher responded

Almadía says it cross‑checked the manuscript with the author’s earlier drafts, notes and email exchanges. After that review, the publisher decided to stand by the prize and the author, stating that the detection tool alone should not dictate publishing decisions. The French house Gallimard, which also uses Pangram for its own submissions, plans to start screening manuscripts with the tool next year, showing that some publishers still see value in the technology despite its controversy.

Wider concerns about AI detectors

In the United States, the Authors Guild warned that AI detectors like Pangram should not be the sole basis for employment or publishing decisions. The Guild notes that Pangram performed well in its own internal tests at correctly labeling human‑written text, but the organization stresses the risk of over‑reliance on a single tool. The debate echoes earlier incidents, such as the confirmed AI‑assisted poem by Thélyson Orélien and disputes surrounding the Caine Prize and writer Jamir Nazir, who retained his award after similar accusations.

Why it matters

For readers, the dispute shows that a high AI‑detection score does not automatically invalidate a book’s literary merit. For writers, it underscores the need for clear standards before AI tools are used to judge authenticity, especially for authors from under‑represented groups.

Was this article useful?
Share

Read next

Oossa · Newsletter

The week in AI, explained

Every Monday: the stories worth knowing, in plain language. Free, no spam.