Who Writes the Caribbean's Story?

The Brief

• Trinidadian writer Jamir Nazir won the Caribbean regional category of the 2026 Commonwealth Short Story Prize for "The Serpent in the Grove."
• Pangram, an AI detector, has flagged three of the five regional winning stories, including Nazir's, as machine-generated. A second detector, GPTZero, classified Nazir's story as entirely human.
• Reports have also alleged that Caribbean judge Sharma Taylor used AI to write the praise blurb that accompanied Nazir's win.
• Nazir denies using AI. The Commonwealth Foundation has stood by him while promising a full review of its policies.
• Kevin Jared Hosein, a former Caribbean regional and overall winner of the prize, has publicly called the prize "dead."
• The detection evidence is genuinely contested. The bigger question it raises for Caribbean writing has to be answered.

Last month, the Commonwealth Foundation announced the regional winners of its 2026 Short Story Prize – the most prestigious annual award open to writers across the fifty-six countries that once made up the British Empire. Of the 7,806 entries submitted, five winners emerged from successive rounds of cuts, one for each of the regions the Commonwealth divides itself into. The Caribbean's winner was a Trinidadian. His name is Jamir Nazir, and his story is called "The Serpent in the Grove."

Read the opening lines on Granta, where the prize winners are traditionally published:

They say the grove still hums at noon. Not the bees' neat industry or the clean rasp of cutlass on vine, but a belly sound – as if the earth swallows a shout and holds it there.

It is, on the page, exactly the kind of writing the Commonwealth's judges praise: textured, atmospheric, carrying a sense of Trinidad's rural interior that anyone who grew up in the central plains will recognise instantly. The Caribbean's judge, Sharma Taylor, described the language as "sublime – precise yet richly evocative – conjuring vivid, lush imagery with remarkable economy."

This should have been a small triumph. For a country of 1.4 million people that has long struggled to keep its newer voices in the international conversation, a Trinidadian winning the Caribbean regional Commonwealth prize is exactly the kind of moment we should be celebrating. Instead, the story has become something else.

Within days of the announcement, an allegation surfaced. The story, several readers argued, read more like the work of a large language model than that of a human. The complaint came first from Nigerian writer Chiemeziem Everest Udochukwu, who flagged what he called "syntactical tics" in the prose – particularly a construction common to ChatGPT-style outputs that one critic described as the "not-X-but-Y" pattern, the kind of sentence that says "not the bees' neat industry but a belly sound." Other readers pointed to similes they described as decorative rather than load-bearing. Ethan Mollick, the Wharton professor who has spent the last three years studying generative AI in professional contexts, ran the story through Pangram, one of the more respected AI detection tools. The result came back at 100 percent machine confidence. Mollick's verdict was widely reported: "In a Turing Test of sorts, it looks like a 100 percent AI generated story just won the Commonwealth Prize for the Caribbean region."

The detection question has also broadened in the weeks since. Pangram ran all five regional winners through its system, and three of them came back flagged – Nazir's story, Malta winner John Edward DeMicoli's "The Bastion's Shadow," and Indian winner Sharon Aruparayil's "Mehendi Nights." Aruparayil has denied the allegations. DeMicoli, at the time of writing, has not formally responded. Whatever this turns out to mean in the individual cases, the structural point is hard to step around. Three of the five stories shortlisted as the best unpublished short fiction in the Commonwealth this year have been flagged by the same detector. That is no longer a one-writer problem.

Nazir has denied the allegations. In a statement posted to LinkedIn, he wrote: "To be pellucidly clear: this work was entirely written by me, drawn from childhood memories of growing up in rural Trinidad." He challenged the reliability of the detection tools themselves, arguing – correctly, on the public evidence – that AI detectors are known to generate false positives on highly polished human prose. On May 22, the Commonwealth Foundation issued a statement saying it took the allegations seriously, had reviewed the evidence, and would continue to support all of its 2026 regional winners. It promised a fuller review of its policies for future cycles. Granta, the magazine that publishes the regional winners on its website, has kept the story online with a notice attached saying that until "definite evidence comes to light," it will leave the piece up.

Granta's response was more remarkable still. Publisher Sigrid Rausing confirmed she had run "The Serpent in the Grove" through Claude.ai to test whether it had been written by an AI. The reading came back nuanced rather than binary: the story might have a "human core" that had been "elaborated around" by AI. That is a different claim from straight machine generation, and it points to where this controversy is actually headed – not a binary, but a spectrum of human-AI collaboration that prize cultures have not yet learned to evaluate. Rausing's own summary tells you most of what you need to know:

"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."

Kevin Jared Hosein, the 2015 Caribbean regional winner of the same prize, and the 2018 overall winner, has been less diplomatic. He has called the prize "dead," described the Foundation's continued support of Nazir as "the second blow," and called for a public apology to all the writers who entered in 2026. His critique of "The Serpent in the Grove" itself is unsparing: "purely nonsensical" in parts, "marketing-speak," bearing "disappointing linguistic homogeneity." That is a respected Trinidadian novelist saying, in public, that a Trinidadian winner of the prize he himself won twice has produced a body of text the Caribbean literary establishment will not stand behind. Hosein also alleges that Nazir's author headshot is itself AI-generated, and that Nazir has acknowledged this. "If he is willing to dupe the Foundation with this," Hosein asked, "why stop there?"

I am certainly not going to call Jamir Nazir a fraud. I do not know whether "The Serpent in the Grove" was written by him, written by a machine, or written by some combination of the two. Neither does Pangram. Neither, frankly, does Ethan Mollick, who has been careful in his subsequent posts to note that 100 percent confidence from a detector is not the same thing as 100 percent certainty. Neither does the Commonwealth Foundation, which is why its statement is on the fence. The only person in the world who knows for certain is Nazir himself. And while there are details in his digital footprint that make readers raise an eyebrow – a thin literary back-catalogue for a writer in his 60s, a self-published 2018 poetry collection, what one observer called a frequent enthusiasm for AI on his LinkedIn page – none of those details prove anything. People come to writing late. People keep low online profiles. People who are excited about AI are not, by definition, AI fraudsters.

On the back-catalogue point in particular, literary prize history runs against the suspicion. Keri Hulme won the Booker for The Bone People, her debut novel, in 1985. Arundhati Roy won the same prize for The God of Small Things, her first, in 1997. Douglas Stuart won it for Shuggie Bain, his debut, in 2020. Jhumpa Lahiri's first collection, Interpreter of Maladies, won the Pulitzer in 2000. Marlon James, whose later Booker for A Brief History of Seven Killings anchors the modern Caribbean canon, was rejected by seventy-eight publishers before his first novel was even printed. Late starters and quiet writers winning big prizes is not a red flag. It is one of the things prizes exist for, and the Commonwealth Short Story Prize most of all.

There is a tougher question we must engage with, because it is the one Caribbean writers and Caribbean readers need to think about now, before the next prize cycle: What does it cost the Caribbean's small, hard-won literary culture if the next genuinely brilliant Trini writer with a polished voice wins a prize and immediately has to defend themselves against an algorithm? "Precise yet richly evocative." "Lush imagery." "A melodic voice that lingers." These are the kinds of phrases the international prize circuit has trained itself to use when it praises Caribbean writing, and they are also fair descriptions of the prose of a dozen Caribbean writers I can think of off the top of my head – writers who learnt their craft in workshops, on long reading lists, on years of revising the same paragraph until it earned its place. Polish, in the literary sense, is what those writers worked for. If polish is now the marker of suspicion, the Caribbean writers most damaged by this case are not the ones who used AI. They are the ones who did not, and who write the way they have always written, and who now have to prove a negative in front of a global audience equipped with a tool that returns binary verdicts on prose it was never quite designed to judge.

That is the part of this story I find genuinely difficult. Victoria Livingstone, a writer I have come to trust on these questions, made the point on her Substack that many of the readers who pointed to "AI tells" in Nazir's prose had already seen the Pangram label before they started reading. The label primed the reading. Once you are told a piece is AI, the not-X-but-Y construction looks like a tic instead of a stylistic choice. The slightly decorative simile looks like a giveaway instead of an aesthetic decision. We end up doing what Livingstone called outsourcing our taste to the machine – letting a detector tell us how to read.

There is a further wrinkle here that turns the argument inward in a way I cannot ignore. Reporting by WIRED and Global Voices has noted that Sharma Taylor – the Caribbean region judge, who praised Nazir's writing as "sublime" – has herself been accused of using AI to write the praise blurb that accompanied his win. Pangram flagged her words too. Whether this allegation is true or not, the suggestion alone is the most cutting version of the argument I am making in this post. The people in the room may no longer be able to tell where their own writing ends and a machine's begins. If the judges' own prose is being run through detectors and coming back flagged, the cultural authority that prize-giving is supposed to confer is in genuine trouble.

I have my own reason to be sceptical of the detectors. A few months ago, out of curiosity, I ran several pages of a Tom Clancy novel – a writer who, whatever else you say about his prose, is unmistakably human, prolific, and pre-AI – through one of the same class of tools. It came back over 75% confident the text was machine-generated. Make of that what you will. The detector industry is fast-moving and the better tools are improving, but a literary culture that defers to any of them as a single source of truth is borrowing a problem it does not need. Pangram is good. It is also fallible. And the day a literary prize jury delegates its judgement to a detector is the day the prize stops meaning anything regardless of who wrote what. And here is another relevant point: While Pangram returned its 100 percent verdict on Nazir's story, GPTZero, another widely used detector, classified the same story as entirely human. Two professional tools, the same text, opposite verdicts. The Foundation's director-general Razmi Farook has been candid about what this means: the institution, he said, has to operate on a "principle of trust" because the detection technology cannot, on its own, settle the question.

All of which hits hardest in the Caribbean, because the Caribbean can least afford it. We have spent decades arguing, against London publishers, New York reviewers, and prize juries that have never set foot in the West Indies, that our voices are serious, that our register is not "exotic" but specific, that the way we describe a grove or a savannah or a fishing village is doing the same kind of literary work as a writer in Yorkshire describing the moors. Now, on top of all that old labour, we have to add a new defence: that we are not machines.

There is a connection here to the post I wrote recently, The Language We Leave at the Door, about the Jamaican opposition MP who was shut down for opening a parliamentary speech in Patois. That post asked who gets to be heard in the formal rooms of our public life. This one is a strange mirror of that question, asked of the formal rooms of our literary life: who is doing the speaking, when the speaking gets through? The two questions are joined at the hip. Both are about voice and authenticity. Both are about the small, structural ways the Caribbean continues to be told that what it sounds like is somehow not quite right. In Kingston, the issue was the official register being too narrow to admit the actual register of the country. In Granta, it is the opposite: the writing reads as polished enough to win, and the polish itself is being read against the writer.

To be fair, the Foundation's position is the right one for the moment. It does not have the evidence to strip a prize, and stripping a prize on the strength of an AI detector that nobody fully understands would set a worse precedent than the one it would correct. Granta is right to keep the story online for the same reason. The bar for taking a writer's work down has to be higher than a probability score. What both bodies have signalled, though, is that the rules will change for the next cycle. They need to. Other major literary prizes are doing the same. The Booker, the National Book Award, several smaller ones are quietly tightening their authorship declarations, requiring writers to disclose any use of generative tools and to confirm, on signed declaration, that the submitted work is substantially their own. The Commonwealth Foundation will get there. The real question is what the verification process actually looks like, and how Caribbean writers, many of whom do not have the institutional support of a literary agent or a publisher's legal team, will navigate it.

Two things can be true at once. The question of whether Jamir Nazir wrote his story is unresolvable by anyone other than Nazir himself, and the public is owed humility about that. The broader question of AI authorship in literary prizes, on the other hand, is here, and it will arrive most painfully at the doorsteps of small literary cultures with the least resource to defend themselves. The Caribbean needs to start that harder conversation now. Not next year. Not after the next prize cycle.

The grove still hums at noon. Whether it hums with the voice of a man from rural Trinidad, or with the voice of a machine trained on a thousand other men from a thousand other groves, is a question that may never be answered to anyone's full satisfaction. The question of how we, as readers, decide which voices to trust, and how we, as writers, hold open the space for the genuine voices that are coming, is one we can answer. We should not let an algorithm answer it for us.