The serpent in the code: What allegations of AI usage in the Commonwealth Short Story Prize mean for regions like the Caribbean

Where to begin? Those could have been the very words that went through Trinidadian Jamir Nazir’s head when he decided to write “The Serpent in the Grove,” his submission to the 2026 edition of the prestigious Commonwealth Short Story Prize. The piece was judged the regional winner for the Caribbean and, as a result, became a contender for the overall prize — except that there’s a huge debate over whether he wrote it at all. On Monday, May 11, I received an alert that Nazir had topped the slew of Caribbean entries, with a note that the news was embargoed until that Wednesday. Exactly a week later, a North America-based colleague emailed me to say he’d seen allegations circulating on Bluesky that the story was the work of artificial intelligence (AI): In a Turing Test of sorts, it looks like a 100% AI generated story just won the Commonwealth Prize for the Caribbean region “for its lyrical precision and haunting atmosphere, the story stood out for the confidence and restraint of its voice.” Published in Granta: granta.com/the-serpent-… — Ethan Mollick (@emollick.bsky.social) 18 May 2026 at 13:50 Mollick’s theory quickly gained traction — he’s a University of Pennsylvania professor, after all, specialising in entrepreneurship, innovation and AI — but it hadn’t yet been reported in any newspaper of record, though I found similar threads emerging on Reddit and X: Well, this is a first: a ChatGPT-generated story won a prestigious literary prize (The Commonwealth Prize). “Not X, not Y, but Z” sentences everywhere, the “hums” trope, and plenty of other obvious markers of AI writing. A major milestone for AI, at any rate…@GrantaMag https://t.co/BWGBpRasNz pic.twitter.com/U6jWejprFv — Nabeel S. Qureshi (@nabeelqu) May 18, 2026 A couple of calls to trustworthy sources revealed that the accusation might have legs. Neither the Commonwealth Foundation nor Granta, where the winning entries are published, had yet made official statements, so I waited. It was an interesting story, but if it ended up going somewhere, I wanted to be able to cover it in a way that wasn’t simply amplifying the noise. The sins of syntax What concerned me at this stage was that the finger-pointing seemed largely rooted in syntax; the conclusion being that a proliferation of dashes, coupled with the approach of describing things as being “not this, but that,” meant that the story must have been generated either wholly or partly by AI. Chat GPT, in particular, is heavily trained in this approach, but this type of writing existed long before Large Language Models (LLMs) did. In an auspiciously timed episode of the podcast This Week in Tech, futurist and author Amy Webb admitted that she neither thinks nor speaks in complete sentences: “I’m just a giant run-on […] this is my natural way of speaking and writing. And the bots out there got hooked on Miranda and now […] I have to convince people like, no, this is just the way that I write.” It got me thinking about the way Caribbean people speak, meandering down long, winding trails of thought before they get to the point. Being of the opinion that AI has done the dash dirty to some degree, the podcast exchange made me consult my always-within-arm’s-reach copy of Strunk and White’s “Elements of Style,” which advises that dashes as a whole — there was no distinction made between en or em — should be used “only when a more common mark or punctuation seems inadequate.” In other words, apply sparingly. In a story of 3,385 words, Nazir used dashes approximately 28 times. Is that too much? And if it is, do we mean “too much” in the sense of overkill, or do we mean it as a reliable gauge of AI usage? What’s the limit, and who gets to decide? Via a WhatsApp call, I asked this of Jarrel de Matas, an assistant professor at the University of Texas Medical Branch’s Department of Bioethics and Health Humanities. His feeling is that because AI detectors can misrepresent, clear disclosure parameters are needed. “How much LLM output can you actually use before it crosses over into irresponsible usage?” he asks. “If you use it for spell check or grammar correction, these are low-level writing mechanics. If it’s just for outlining, researching…I think those are acceptable ways to use it, but they should be disclosed. But when you are engaging with a system that you are asking for ideas, you’re giving it some kind of authority over the creative process.” He calls it “a watershed moment for creativity.” Using AI to detect AI By May 19, the UK Guardian ran a story about the controversy, which quoted Granta’s Sigrid Rausing, as saying, “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.” The magazine stated that it had put Nazir's writing through Claude to determine whether his story had been AI generated, which is a classic case of “himself to himself,” to quote The Mighty Spoiler. The chatbot concluded that it was “almost certainly not produced unaided by a human.” Trinidadian writer Kevin Jared Hosein, the 2015 regional winner of the prize, was livid that the Commonwealth Foundation continues to stand behind Nazir, who insists “The Serpent in the Grove” is his own work, and that the story — along with other 2026 entries that have also been flagged for their potential use of AI — remains on Granta’s website. Rausing’s admission that there is “a certain irony in the fact that beyond human hunches AI itself is the most efficient tool we have for revealing what is AI generated,” likely brings little consolation. Hosein challenged, “You have human eyes. You will be the jury. Jamir’s headshot, first of all, is a product of AI modification. He has admitted this. If he is willing to dupe the Foundation with this, why stop there?” Hosein is one of the judges for the 2026 Brooklyn Caribbean Literary Festival, which just posted its policy on AI use for its short story contest. The parameters prohibit the use of “LLM-based tools to generate, edit, revise, summarize, or otherwise assist in preparing your submission.” While some regional writers were pleased with the festival’s stance, it seems a bit idealistic considering you can’t even google something now without being met with the search engine’s standard AI Overview at the top of the page. De Matas reminds me that when ChatGPT burst onto the scene in November 2022, the talk was about the death of the English major. “At that point, the threat was that we wouldn’t think anymore,” he recalls. “But now, it seems the conversation has shifted to, ‘We cannot be creative anymore.’” The competition’s judges have come under their fair share of fire. A WIRED article noted that Jamaican author Sharma Taylor, who sat on the jury, “has been accused of using AI to craft her descriptive blurb that accompanied the listing of ‘The Serpent in the Grove’ as a regional winner,” adding that Pangram — which the AI experts who raised the alarm in the first place employed to evaluate Nazir's story — had assessed Taylor’s text as “AI-assisted.” Between a rock and a hard place Both the Commonwealth Foundation and Granta have received criticism for the equanimity of their response, but really, how else could they have played this? “We are making assumptions to fill a gap,” De Matas explains. “I think that certain ‘obvious markers’ make parts of the text read as AI-generated, and I also think that it’s difficult to prove.” He’s interested in how we can use this moment to develop tools that can ascertain, without a doubt, what many people believe they’re seeing, and suggests this can be accomplished by tracking trends — data sets, stylography: “A lot of work would have to go into coding style. It’s not impossible, but we’d need incentives to create the guard rails.” As a machine learning model, he continues, Chat GPT trains on the data it finds and continuously refines it, though it never really loses the structure. “You can ask it a hundred things and it will say 99 in the same way,” he adds. “It doesn't change the style. It changes the information.” Against the backdrop of a quickly advancing AI race, such a tool seems critical, especially since these developments appear to push us towards more usage. “We think we have the agency,” De Matas laments, “but we don’t; everything is becoming AI-driven. It’s affected image generation, the smartphone, operating systems and so, we’re doing less and less.” Rise of the bots? A friend recently recounted a discussion she’d had with a Trinidadian secondary school student, who told her that any School-Based Assessments (SBAs) she and her peers submit are put through an AI-checker, and there’s a certain percentage of AI detection they are not allowed to surpass. Fair enough, but here’s the rub: if she submits a “perfect” paper, her work is judged to be AI. She’s then forced to submit papers with deliberate errors, for which she gets marks docked. In a LinkedIn post, Nazir took a similar line of defence, questioning the accuracy of AI detection tools like Pangram: “It is a fact that AI detectors are completely unreliable generating false positives when evaluating highly polished human prose. Algorithms cannot be allowed to dictate that a human is incapable of producing a high level of literature.” So what constitutes great writing? To shed some light, I revisited Plato’s “Phaedras,” which mentions things like the power of truth to tap into the universality of the human experience. In the text, Socrates, who values rhetoric more than the written word, declares, “[T]his discovery of yours will create forgetfulness in the learners’ souls, because they will not use their memories; they will trust to the external written characters and not remember of themselves […] you give your disciples not truth, but only the semblance of truth, [having] the show of wisdom without the reality.” Plato, clearly a soothsayer, might as well have talking about artificial intelligence. As AI continues to be fed human content, it will get better; it already has. This makes it all the more difficult to discern what is human and what isn’t. An educator friend posed a theoretical question to me in the midst of all the furore: “If a good story is proven to have been written by AI, does it make it any less of a good story?” My personal feeling is that yes, it does, because it lacks the authenticity of the human experience, which is where the power and truth in stories inherently lie. It reminds me of a yarn about a purported face-off between famed Shakespearean actor Richard Burton and American Catholic bishop Fulton Sheen. In the (admittedly AI-esque) story, they were each asked to read Psalm 23 to see who could evoke the more emotional response. Burton’s rendition earned him a standing ovation; Sheen’s, contemplative silence, with the judge determining that Burton knew the passage, but Sheen knew the shepherd. De Matas and his peers now routinely come up against the likelihood that AI is changing how we engage with our own humanity — “our processes of creating connections, originality, how we derive inspiration, things we took for granted because we were human.” Even Pope Leo XIV has weighed in; his first encyclical argues that humans flourish because of limitations, not in spite of them. AI, on the other hand, makes everything immediate and easy. There is no struggle and, as any writer can attest, the struggle is real. “I think what we are losing is the self, the ability to be honest with each other; with ourselves,” De Matas stresses, “and that is such a nebulous concept, we can’t quantify it. We can’t code honesty.” Literary backlash Recognised writers have not deemed Nazir’s story worthy. Despite the Commonwealth Foundation’s assessment of its “lyrical precision” and “confidence and restraint of its voice,” Hosein found that it “lacks intentionality behind the strangeness it evokes as ‘great’ literature. [N]one of his metaphor and simile serve the characters or the narrative. The closer you look at it, you may see how it employs heavy use of (unearned) aphorism and I always find AI stories to be floaty or glide-y, like it can never really settle on a particular moment and sustain emotion or tension, only ideas of it.” Jamaican Booker Prize winner Marlon James simply quipped, “Forget AI for a second. A story won an international Competition with a line like this: ‘The girl smiled like sunrise over a sink.’” Such comments point to a two-pronged issue. First, whether Nazir and the other two writers accused of using AI were truthful when they declared to the Foundation that their work was of their own creation, and second, whether the writing has merit. Addressing the first point in a May 19 statement, the Commonwealth Foundation defended its “robust” judging process: “Until a sufficient tool or process to reliably detect the use of AI emerges that can also grapple with the challenges pertaining to working with unpublished fiction, the Foundation and the Commonwealth Short Story Prize must operate on the principle of trust.” As criticism mounted, however, the Foundation’s Director-General Razmi Farook felt it important to address the organisation’s “duty of care.” Calling AI “the single biggest issue facing much of the creative world,” she noted, “[W]hile we welcome constructive debate surrounding this complicated and nuanced matter, we are deeply concerned by the tone of much of the discourse surrounding it. At the centre of this are real people: striving writers, dedicated judges, and a passionate team.” She further noted the Foundation’s disappointment that the work of not only current but previous winners of the prize had been called into question, but added, “We feel it is vital to recognise the upset felt by all those involved in or affected by the recent discourse surrounding the Short Story Prize. We offer our support to our winners, shortlisted writers, judges, alumni and wider community.” Their message is clear, at least to me, on the point of calibre: writers, readers, critics, AI specialists and social media commentators can all have their opinions, but the final word on whether a piece is deserving or not lies with the people who have been entrusted to judge it. For De Matas, this is precisely the point: “In the same way that AI tools are not transparent, and AI developers are not transparent, this moment should be an eye-opener for the prize itself, because they aren’t exactly transparent either. It says as much about the creative output as it does about the basis for judging those outputs.” Is there a way to tell? AI was always going to become an integral part of our existence. Technology has been going in that direction for quite some time, but the hope was that it would — like Rosie in “The Jetsons” — competently perform operational tasks for us, leaving humans to concentrate on creative ones. Instead, we’re being faced with Open AI’s Sam Altman boasting about “a new model that is good at creative writing” and “got the vibe of metafiction so right.” While AI models are proceeding at breakneck speed, even as they negatively impact both planet and people, regulation, legislation, and discernment tools are lagging behind. On Substack, James O’Sullivan posited that stylometry, “a technique within the computational humanities which sets out to measure literary style — i.e., how a text is written, or indeed, by whom it was written,” could be useful in cases like this. “Count up the most common words in a corpus (which are always function words like ‘the’, ‘and’, ‘of’, ‘I’, ‘was’),” he advised, “standardise those counts against the corpus as a whole, and the resulting profile turns out to be a reasonably reliable identifier of who wrote what.” In another blog post, O’Sullivan agreed that “The Serpent in the Grove” had a few red flags: “Negation staged as the setup for revelation is among the most reliable rhetorical defaults of current language models when asked to perform a literary register.” He also mentioned various aphorisms that have “the cadence of profundity but little real meaning,” “deliberately strange verb-noun pairings,” and “the reflex toward parable in place of incident.” “But maybe,” he continued, “Nazir did write The Serpent, which wouldn’t be the first piece of formulaic fiction to reach award-winning status (I can honestly see ChatGPT receiving a Booker Prize someday).” Bringing the discussion around full circle to the tweet that started the bot-text ball a-rolling, O’Sullivan said, “The framing here is telling. By calling it a Turing test, Ethan Mollick implicitly conceded the central problem that readers (and publishers) face in the age of LLMs: whether a piece of prose was written by a human can no longer be reliably determined through reading alone. If it could, well then no test would be needed.” He then cut to the heart of the matter: “The consequences of this reach far beyond Nazir. If AI ‘detectors’ and Bluesky accusations are allowed to function as prima facie grounds for a public denunciation of authorship, the writers harmed will be the ones whose stylistic surface overlaps with generative prose for reasons that have nothing to do with AI use.” This includes, from my own editorial experience, writers for whom English is not a first language, academic writing, and flowery prose with an over-reliance on that ubiquitous em dash. Could it also include Jamir Nazir’s “unusual writing process,” which he shared with the UK Observer’s Erica Wagner? “Chronic health conditions which make sustained, desk-bound typing physically impossible,” he told her, require him to write via speech-to-text, with minimal keyboard editing. It puts a very human face on what many are regarding as a primarily technical or even moral issue. “The problem we now face is this,” O’Sullivan surmises. “[W]e do not want so-called authors winning literary prizes for AI slop, but equally, we don’t want a situation wherein the threshold for being publicly named as a fraud is a screenshot from an AI detector and a Bluesky post.” Vestiges of colonialism I’ve heard all manner of questions raised about this debacle, perhaps the most common being, “How could the judges not have known?” Do you know, without a doubt, that Nazir and the two other accused are guilty of using AI? I don’t, but I’ll tell you what I do know. Despite the prize’s judging panel being as diverse as the regions it serves, they often don’t see us. In Lina Abushouk’s article “How to read postcolonial writing,” she argued that the Foundation’s rationale for choosing the story revealed “a critical vocabulary that has become entirely decorative — terms like ‘richly evocative,’ ‘sensory detail,’ and ‘melodic voice’ floating free of any engagement with the actual sentences on the page. Nazir’s selection, therefore, is less about him using AI and more about the irony of Euro-American conceptions about the unintelligibility of postcolonial writing.” She went on to explain that “large language models do not invent anything. [T]hey learn to predict what kinds of sentences tend to follow other kinds of sentences. This means that when an LLM generates ‘literary’ fiction set in the postcolonial Caribbean, it does not reach for originality — it reaches for the most probable version of what such fiction has looked like in the texts it has been trained on. It reproduces the expected atmospheric density, the expected weight of landscape and labor, and the expected imagery of poverty and endurance. The scandal is that the existing formulae for ‘authentic’ postcolonial prose are already so codified that a language model can reproduce them convincingly. In this way, AI does not disrupt literary taste so much as expose its furniture.” Abushouk continued, “Others have pointed out that some of the judges themselves are from the Global South, which would mean it is not a simple story of metropolitan outsiders imposing their expectations. But actually, this would make it something more insidious: a set of aesthetic assumptions so thoroughly institutionalized that they can be reproduced from within.” Comparing the whole affair with Amos Tutuola’s relationship with Faber and Faber in the 1950s, Abushouk surmised, “Faber did not celebrate him despite his linguistic unevenness […] they celebrated him because of it. The ‘wrongness’ of the English was the point — it authenticated the primitive. […] The structure of the error is identical across seven decades: Incoherence that would disqualify a European writer is reframed as authenticity when it is attached to the right cultural geography. This is what the AI controversy ultimately reveals — not a new problem introduced by technology, but an old one made newly legible.” The issue of access The notion of Tutuola being perceived as “ungrammatical” gave me pause. I thought back to Wagner’s interaction with Nazir, and her musings after an AI checker determined his headshot was either heavily edited or AI-generated: “Why wouldn’t you use all the tools available to you if it was hard for you to get a decent headshot done?” She added, “His initial message to me on LinkedIn contained misspelled words, was ungrammatical. What light does that shed on his polished, prize-winning story? Any? Some? None?” Some would say the light is harsh, hot, interrogative; that it unmasks a grifter. Yet, there are perfectly punctuated pieces of writing that fail to say anything valuable, and grammatically imperfect stories that can touch something deep in the hands of a competent editor. If Tutuola were attempting to start his writing career today, would a publisher take a chance, or would he be overlooked on account of his solecisms? Would he use AI as a tool if he felt it levelled the playing field? Is it even a tool? A recent Bluesky thread suggested that it wasn’t in users’ best interests to refer to LLMs as such: Speaking of which: Consider referring to ChatGPT, Gemini, Claude, and so on, as ”products” or “apps” rather than “tools”; the latter tends to obscure their relationship to the companies behind them and makes it sound like they exist just to help us. (3/x) — Vauhini Vara (@vauhinivara.bsky.social) 12 May 2026 at 12:33 Terminology aside, De Matas says that using AI as a tool (as opposed to some kind of creative collaborator) allows people to transcend linguistic barriers; if that offers an advantage, then is it an advantage in expression, or imagination? “It’s changing the rules of authorship; the nature of creation,” he confirms. “We have relied on a traditional idea of authors associated with pedigree, and what these AI systems are showing is that a lot of us can become ‘artists.’ Is this just another evolution of that? The bigger question is, where does authorship now need to evolve?” In his recent experience of judging medical residents on case reports they had to present in person, De Matas found the quality of some of the presentations did not match the level of writing. Proving human authorship, therefore, might well have less to do with the writing and more with “the connections and authenticity they bring to it.” As with every thesis hearing in history, if you can articulately defend your work, chances are you authored it. Either way, literature has gatekeepers and competitions like the Commonwealth Short Story Prize help unlock the gate. Naturally, writers must put in the effort to hone their skills and develop their voice, which is why festivals like the Bocas Lit Fest are so crucial. Bocas has consistently offered regional writers all the support it can muster, even as its own support systems have pulled away. It’s not just writers, though; it’s readers too. If we have any hope of keeping up with AI, reading is non-negotiable. Start with books that were created before AI became a thing. They will feed discernment, expand vocabulary, nourish the spirit, spark the imagination, and make us better (albeit not wholly) equipped to separate the craft from the chatbot. Of course, we’ve probably already crossed the Rubicon. Whether Nazir did or didn’t use AI, or to what degree, the doubts cast over the integrity of the work, justified or not, have likely stymied his writing career before it even got going. He has become the AI-generated face of an issue that is already proving to be bigger than us. To repurpose a line that Bocas festival director Nicholas Laughlin wrote in an obit for V.S. Naipaul, “The moral algebra of art is difficult, and it should trouble us.” The human implications of AI’s moral algebra should disturb us even more.