“The Serpent in the Grove” was a story that I read twice. On the first reading, I looked for evidence. In the second attempt, however, I just tried to enjoy it as a story. Both readings were quite different experiences. Jamir Nazir’s short story won the Caribbean regional prize at the Commonwealth Short Story Prize. Published in Granta, it soon enough became the centre of a furore over whether a human being actually wrote it. AI detection platform Pangram identified it as 100% machine-generated. former Palantir employee posted about its “obvious markers of AI writing” on X. Ethan Mollick, a professor at the University of Pennsylvania, called it “a Turing test of sorts.” The internet, as it does, ran with it.
Also read: The Musk vs OpenAI verdict answered nothing about AI governance
So, I searched for evidence to support the case made against the author. As I expected, I found plenty of “not X but Y” constructions: “Not the bees’ neat industry… but a belly sound.” “Not gratitude, exactly. A fact.” “Not like a mother, like a judge.” Aphorisms fly at you one after another: “Water is jealous.” “Shame is a substance.” “Doing is a treacherous bridge.” The end of the story features anaphoric repetitions: “The grove remembered. The house remembered. The boy remembered.” Tick, tick, tick. These are classic AI techniques, critics exclaimed. They are not wrong.
But there is one catch. If it weren’t for the scandal around this story, I probably wouldn’t have read it. And when you go into a piece of writing already holding a checklist, you will find what you are looking for. Every stylistic tic becomes a data point. Every metaphor becomes suspect. The story stops being a story and becomes a specimen.
Without taking the list into account, “The Serpent in the Grove” is a story written in a Trinidadian dialect. The Trinidad dialect is not the normal AI-Caribbean – “It had a well there once, and a woman,” “Is Marsha!” – rather, it is very much placed and particular. The idea of structuring a story by which the biblical serpent from the garden is no longer a temptress, but the man, and the vine that allows the woman to escape is nothing more than her own willpower is definitely not an idea that comes randomly.
Also read: The Monet AI experiment that exposed how we think about creativity and authenticity
Nevertheless, doubt still lingers. Granta publisher Sigrid Rausing, to her credit, didn’t pretend otherwise. “It may be that the judges have now awarded a prize to an instance of AI plagiarism – we don’t yet know, and perhaps we never will know,” she explained. Moreover, she also said that she used Claude to verify if the story was written by a person or by AI. Claude, which is an AI chatbot itself, said the story was probably not pure AI, but probably not entirely human either.
Which raises the obvious question: if an AI cannot reliably detect AI writing, and humans demonstrably cannot either, what exactly are we working with? The Commonwealth Foundation said it doesn’t use AI detectors during judging because submitting unpublished fiction to one raises concerns about consent and artistic ownership. So their process runs entirely on the honour system – writers self-declare that they wrote their work. Director General Razmi Farook put it plainly: the foundation must “operate on the principle of trust.”
Trust. In 2026. When the tools we have built to replace trust are also the tools we are using to check whether trust was broken. This is the corner the literary world has backed itself into. AI detectors are unreliable, with a documented tendency to flag non-Western English as suspicious, which means a Trinidadian writer working in Caribbean oral cadences is at a structural disadvantage before anyone reads a word. Human judges, trained on a Western literary canon, carry the same bias in a different form.
None of this proves Nazir wrote every word himself. None of it proves he didn’t. That’s exactly the problem. Literary prizes are now running a silent Turing test with every submission, and they have no instrument to grade it. The best AI detectors available also returned a verdict of maybe. The judges, experienced literary professionals, read the story through multiple rounds and gave it a prize. The internet read it in an afternoon and decided it was slop. I read it twice. I’m still not sure. And that uncertainty isn’t Nazir’s problem to solve. It is ours.
Also read: 5G network slicing explained: The tech behind Airtel’s Priority Postpaid