Caine prize winner denies AI use after detector flags story

A win turned into a scandal in five days flat. Simbarashe Kundizeza's story The Blue Fighter took the Caine prize for African writers on 6 October. Five days later, The Atlantic ran a piece by journalist Vauhini Vara reporting that she'd put the story through the AI-detection tool Pangram — software that itself uses AI — and it came back almost completely flagged as artificially generated. The only part Pangram didn't flag, per Vara, was an epigraph quoting an unpublished poem by South African writer Vangile Gantsho.

Kundizeza isn't having it. In a statement on X, he said: "For the avoidance of doubt, I wrote 'The Blue Fighter'. I did not use generative AI to write it. It is entirely my own work." He's also pushing back on Vara's observation that a 2021 story of his reads nothing like this one, stylistically — pointing out, fairly, that writers are allowed to grow and change over five years.

The Caine prize is now formally reviewing the allegations. Its chair, Ellah Wakatama, didn't mince words with the BBC's Front Row: she's against a detection tool being treated as "the final arbiter," and she said she isn't willing to let black writers in particular become "the testing ground for something that is at best still rough and ready."

Here's where it gets genuinely murky rather than one-sided. Vara cited Pangram's own research showing false positives are rare — including for non-native English speakers — plus a peer-reviewed study finding "strong accuracy" on mock academic papers written with varying degrees of AI help. But she also told the BBC there's "relatively little third-party research" on how accurate AI detectors really are. So you've got a tool with some real evidence behind it, and also real gaps in independent verification. Both things are in the record. Neither cancels the other out.

This isn't isolated, either. In July, Trinidadian writer Jamir Nazir won the Commonwealth Foundation's short story prize and immediately faced social-media claims his story showed AI "markers." Last month, Canadian-Haitian author Thélyson Orélien's novel was pulled from the Goncourt longlist over similar AI allegations — a separate case entirely, and worth knowing Orélien has also faced accusations of plagiarism in earlier, pre-AI-era work.

What to do with this if you're an author entering prizes: know that a detection score is not proof, and it isn't nothing either — treat it the way you'd treat any single piece of contested evidence. If your work ever gets flagged, do what Kundizeza did immediately: put a clear, specific denial on the record, fast, before the story calcifies without you in it.

Watch for the Caine prize's review outcome — that's the next real checkpoint here, and there's no date yet for when it lands.