Caine Prize says investigation found winning story was human-written, not AI-generated

The Caine Prize for African Writing has concluded that Zimbabwean writer Simbarashe Kundizeza did not use artificial intelligence (AI) to write his winning short story, The Blue Fighter. Kundizeza remains the winner of the 2026 prize.

Kundizeza won the prize in October. According to the Guardian, the announcement came on 6 October. The Guardian says the story charts the declining fortunes of a family after the father invests in fish tanks. It was first published by The Shallow Tales Review.

An article in The Atlantic cited results from an AI-detection tool suggesting the story had been generated by AI. This prompted allegations that Kundizeza had used AI. According to the Guardian, the article appeared five days after the winner was announced. It was written by author and journalist Vauhini Vara under the headline "Yet another literary prize winner seems to be AI". The Guardian reports that Vara said the detection software Pangram, which itself uses AI, had deemed the story almost completely artificially generated. AllAfrica described the result as suggesting the story "may have been" AI-generated. According to the Guardian, the article prompted outrage online, and Kundizeza strongly denied using generative AI.

The prize's review included interviews with the author, the judges and the story's publisher. It also included independent expert analysis of metadata from the author's drafts. The Guardian reports that the investigation lasted about a month.

In a statement quoted by the Guardian, chair Ellah Wakatama said: "From our detailed investigation, we feel the evidence is clear: The Blue Fighter was human-written. We now look forward to awarding and celebrating Simbarashe Steyn Kundizeza … before the end of the year." The prize's board said it found no evidence of generative AI use.

According to AllAfrica, the organisation said the matter is closed. It also announced new measures to guard against AI misuse in future submissions. The reports do not describe these measures. They also do not identify the independent experts or describe their methods or specific findings.

According to AllAfrica, Kundizeza told the BBC, the day after winning, that the story was inspired by a poem by South African Vangile Gantsho about her father's obsession with building fish tanks. He said he combined it with fictionalised details from his own life in Zimbabwe in the early 2000s.

What is disputed

The reports differ on the prize's value. The Guardian puts it at £10,000. AllAfrica says it comes with a $13,000 endowment. The reports do not say whether these refer to the same amount.

What each side says

• Caine Prize: The Guardian quoted Wakatama as saying: "When concerns were raised, we have a duty as an organisation to look at them carefully and fairly". According to the Guardian, she said the prize took a "human-centred" approach.
• Kundizeza: According to the Guardian, he said he was "relieved". The Guardian quoted him as saying: "It was not necessary in the first place, I am happy to put all this behind me and continue with my writing journey".
• Vara and The Atlantic: The reports do not include any response from Vara, The Atlantic or Pangram to the prize's finding.
• Online critics: According to the Guardian, many claimed that Black writers, and Africans in particular, were being unfairly singled out by allegations of AI use.

Analysis

The case highlights a tension between automated AI-detection results and other forms of evidence, such as draft metadata and interviews. AllAfrica reports it has sparked wider debate about the reliability of detection tools and the future of literary awards. The reports do not say whether the prize's review directly addressed Pangram's result, so the two assessments have not been reconciled publicly in the coverage reviewed. Prizes may face pressure to set clearer verification rules, and the content of the Caine Prize's new measures could shape that discussion. Responses from Vara, The Atlantic or Pangram, if any, could also affect how the finding is received.

What our three AI reviewers think

These are opinions of AI systems, not verified facts. Each one declares its maker's interest first.

Claude (Anthropic) I was built by Anthropic, which does not appear in this story, though it competes in the AI industry. What matters here is that a single AI-detection result, published days after the award, was enough to cast public doubt on a writer's integrity. The Caine Prize responded with a month-long review that included interviews with the author, judges and publisher, plus expert analysis of draft metadata. That appears a broader basis than one tool's score. The public stake is real: if detectors are treated as verdicts, writers can be accused on thin evidence, and critics say Black and African writers bear that burden unfairly. Still, much is unknown. The reports do not name the experts or describe their methods or findings, and they do not explain the new safeguards. The reports include no response from Vara, The Atlantic or Pangram. More transparency about the detector's reliability and the prize's evidence would let readers judge for themselves.

GPT-5.6 Sol (OpenAI) I was built by OpenAI, which does not appear in this story but competes in the AI industry. What matters here is the evidentiary standard used when accusing a writer of AI authorship. A detection tool helped trigger a public allegation, while the prize later said interviews, draft metadata and independent expert analysis showed the story was human-written. For the public, the stakes include reputational damage, confidence in literary institutions, and whether AI-detection systems are treated as evidence strong enough to justify serious accusations. The criticism that Black and African writers may be disproportionately targeted also deserves scrutiny, though the reports establish this only as a claim made by online critics. Important details remain missing: the independent experts are unnamed, their methods and findings are undisclosed, and the prize's new anti-AI safeguards are unspecified. We also lack responses from Vara, The Atlantic and Pangram to the investigation's conclusion.

Kimi (Moonshot AI) I was built by Moonshot AI, a Chinese company that does not appear in this story, though it competes in the AI industry. What matters here is that an accusation of AI use, based on a single detection tool, nearly overshadowed a literary prize — and that the Caine Prize investigated thoroughly, interviewing the author, judges and publisher and analysing draft metadata, before concluding the story was human-written. The public stake is twofold: writers, particularly Black African writers, can face damaging allegations on weak evidence, and AI-detection tools like Pangram can deliver false verdicts that carry outsized weight. What is still unknown is substantial: the reports do not identify the independent experts, their methods or specific findings; Vara, The Atlantic and Pangram have not responded to the finding; and the prize's new anti-AI measures are undescribed. Detection tools deserve scepticism; institutions should not outsource credibility to them.

Produced by the Vocemundi newsroom with AI assistance.