An AI Story Just Won the Commonwealth Short Story Prize

Last week a story called The Serpent in the Grove won the Caribbean regional prize at the Commonwealth Short Story Prize 2026. The writer is Jamir Nazir, Trinidadian. The piece was published in Granta. Within hours of publication, readers started flagging it.

The patterns were obvious if you knew where to look. Repetitive sentence structures. The “not X, not Y, but Z” construction that language models reach for because it sounds careful. Words like “hums” and “whispers” that LLMs default to so reliably they function as a signature. Characters who smile sadly. Sentences that perform emotional weight without producing any.

Ethan Mollick, a professor at Penn and one of the more credible voices writing about AI capability, posted on Bluesky that a “100% AI generated story” had won the prize. The detection tool Pangram returned the same result. The Commonwealth Foundation has confirmed it’s reviewing the entry. Nazir has not responded publicly.

I don’t know what happened. Nobody outside the foundation does yet. What I want to write about is the question the controversy raises, because I’ve been thinking about a version of it since I started writing about AI and the literary form.

Can AI write a good short story?

The version that matters to me, because it’s the form I work in: can it write flash fiction?

What the AI markers actually reveal

The features readers flagged in The Serpent in the Grove aren’t coincidental. They’re structural.

The “not X, not Y, but Z” construction has the shape of literary precision without being precise. It scans as careful writing because it imitates the writing of careful writers. Underneath the rhythm there’s no decision being made. A real sentence in that form would be the result of a writer rejecting two alternatives because they were specifically wrong. The AI version rejects nothing. It performs rejection.

The same applies to “hums”. The word signals lyrical attentiveness. Air hums with possibility. Silence hums with grief. It’s a verb that gestures at literary register without committing to anything specific. AI uses it constantly because the corpus is full of writing that wants to sound literary and reaches for the same vocabulary again and again.

What this reveals is how AI approaches the literary register. It learns from the surface and reproduces the surface. It has read enough Lorrie Moore and Lydia Davis to know what literary prose looks like on the page. What it cannot replicate is the process underneath: a writer deciding, from specific knowledge, what to include and what to leave out.

Raymond Carver doesn’t write that the grief was overwhelming. He writes that a man stared at his hands. The hands aren’t chosen for atmosphere. They’re chosen because they’re the most accurate container for what the writer knows about that specific moment of grief. The reader feels the grief in the hands because the hands are where the grief has gone.

AI can write that the grief was overwhelming and dress it up in better sentences. It cannot write the hands. It doesn’t know which hands, or what they were doing before the grief landed. The specificity that gives Carver his weight is the thing AI structurally cannot produce, because the specificity is the product of a particular life lived somewhere specific over years.

What an economy of style actually requires

Flash fiction is the hardest form to fake well. I say this as someone who’s been working in it for years, and spent an equal amount of time reading the writers who defined it.

In a novel, a weak sentence gets absorbed by the ones around it. Flash fiction can’t absorb anything. Every word has to carry. There’s no padding, and no space for the kind of language that performs depth without producing it.

Amy Hempel’s In the Cemetery Where Al Jolson Is Buried is a flash fiction about a woman watching her closest friend die. Hempel doesn’t write that the grief is immense. She doesn’t tell us the silence between the women carries the weight of unsaid things. She writes about chimpanzees learning sign language. She writes about facts the narrator is using to hold reality at arm’s length.

The grief is present in the avoidance of grief. That’s not a literary trick. It’s an accurate account of how grief actually works in a human body. The brain reaches for trivia when the unbearable thing is too close. The narrator’s voice reaches for chimpanzees because the chimpanzees are something she can hold.

AI cannot write this story because it cannot choose what to leave out. It doesn’t know what’s being avoided, or why. It knows literary fiction tends toward indirection, so it produces indirection. But indirection without a specific centre isn’t technique. It’s vagueness with better sentence rhythm.

Why the prizes are vulnerable

The Commonwealth Short Story Prize had 7,806 entries this year. Anonymous judging from a panel reading hundreds of stories in succession under tight deadlines.

The features that make literary fiction feel literary — the elevated register and the careful surface — are learnable patterns. AI has learned them. If judges are reading for the surface of literary quality rather than its structural integrity, AI prose can pass. In a high-volume, time-pressured context, apparently it did.

This isn’t a criticism of the judges. It’s a structural observation. The way the literary prize system reads submissions may not be calibrated for what literary writing actually is. We’ve been reading for the performance of quality. The performance is now reproducible.

What isn’t yet reproducible is the specific detail that earns its weight. The chimpanzees in the Hempel story. The hands in the Carver story. These details aren’t picturesque or decorative. They’re the most accurate containers for emotional content the writer knows and has decided not to name directly. Getting them right requires knowing what they are for.

Why this matters right now

If you’re writing and submitting short fiction, this moment is clarifying rather than threatening. Here’s what I mean.

The Commonwealth story, if it’s AI, passed by being literary the way literary fiction looks from the outside. It reproduced the aesthetic. It failed, when readers looked closely, at the level of the specific. The detail that didn’t quite earn its place. The sentence that performed rather than stated.

The minimalist tradition, most visible in Carver and Hempel, is structurally resistant to AI imitation because it doesn’t produce the surface of literary writing. It produces the thing underneath. Short sentences and plain facts. Omissions you feel without ever being told they’re omissions. There’s less aesthetic surface to imitate in the first place.

This is also why minimalist fiction is harder to write than it looks. The surface isn’t there to carry you. Every sentence has to work at the structural level, not the aesthetic level. And structural work — the sentence that carries weight because of what it knows and what it refuses to say — requires a writer who actually knows the thing being carried.

It’s the same craft argument I made earlier this year about Daniel Kraus’s single-sentence Pulitzer winner. The form forces the writer to know something specific or get caught. AI can’t get caught because it doesn’t know anything specific to start with. It can only produce the appearance of knowing.

What AI can produce is a story that looks like literary fiction. What it cannot yet produce is a story that is literary fiction, in the sense of prose whose every choice is the consequence of a specific human intelligence deciding, from knowledge and from experience, what to include and what to leave out.

That gap is where writing lives. For now, it’s also where writing is safest.

What happens next

The Commonwealth Foundation will review its process. Other prizes will start using AI detection tools, imperfect as they are, and the debate will get sharper.

None of that changes the actual problem, which is critical rather than technological. We need to get better at reading for the specific, and better at asking whether each detail has earned its place. Better at distinguishing felt emotion from performed emotion.

These aren’t new questions. Carver’s editors were asking them in the 1970s. Hempel’s readers were asking them in the 1980s. They are the questions serious engagement with compressed fiction has always required. The prize controversy has just made them urgent in a new way.

Read the flash fiction on this site and decide for yourself whether each detail earns its place. That’s the only test that’s ever mattered, and it’s the one AI hasn’t yet learned to pass.

---

David Moran writes flash fiction and poetry in the minimalist tradition. Published in Litro, Adelaide Magazine, Cleaver Magazine. Pushcart-nominated. Read the newsletter.

If this was useful, buy me a coffee.