The Caine Prize for African Writing said Wednesday that Zimbabwean writer Simbarashe Steyn Kundizeza keeps its £10,000 award. In October, journalist Vauhini Vara wrote about his winning short story, The Blue Fighter, in The Atlantic. She said the AI detector Pangram flagged almost all of it as AI-generated. She also pointed to what she called hallmarks of chatbot prose, and wrote that she may well be wrong. Kundizeza denied using AI. For its review, the prize did interviews and had a forensic investigator check his drafts. The prize says he wrote four drafts in Word over as much as 27 hours. "The review found no evidence that the story was produced using generative AI," it said. It also said detectors alone "are not a reliable basis for determining authorship."

The Atlantic Got It Wrong, and Claims Like This Target Black Writers

Many writers say the article never had real evidence, and that this kind of claim keeps landing on Black writers.

The Atlantic never had real evidence. Guardian US senior reporter Julia Carrie Wong called it "the Atlantic's un-evidenced claim" on Wednesday. She said the prize's account "should hopefully serve as a reminder of what counts as evidence... and what doesn't." Nebula Award-winning novelist Aliette de Bodard said Pangram "very clearly returned a false positive." A 2025 University of Maryland study found Pangram wrongly labeled 2% of documents as AI.

Writers saw a pattern from the start. In October, journalist Karen Attiah said it was "really chilling" to see how quickly "Black, Caribbean, and African writers and academics have become the global face of AI use and plagiarism." Nigerian writer and editor Oghenechovwe Donald Ekpeki said "AI is just the tool of the age. Before this, it was plagiarism." Cambridge professor Doyle D. Calhoun told CNN that these claims tend not to "stick" to white writers the same way.

Writers are allowed to change their style. Vara had said the story reads differently from Kundizeza's 2021 story. Nigerian novelist Elnathan John answered, "My writing changes every few years. Sometimes significantly."

Detectors Still Have a Job to Do

Pangram and its backers say AI writing is flooding publishing, and the tool is usually right.

Pangram says magazines are swamped with AI submissions. Pangram CEO Max Spero told CNN in October that literary outlets "are getting flooded with submissions." He said the goal is to "help people understand what's AI and what's not."

Pangram says it only flags writing it's sure about. Spero says Pangram calls something AI-assisted only when it's extremely confident. Pangram told CNN its false-positive rate is about 1 in 10,000, and CNN says two peer-reviewed studies back that up. Vara told Lit Hub in October that Pangram "is very effective," citing Pangram's own research and some outside studies.

Pangram's CEO wasn't satisfied by another prize's review. Earlier this year, the Commonwealth Foundation cleared another flagged prizewinner. In October, Spero told CNN he wasn't convinced by that review. Granta publisher Sigrid Rausing wrote about the same case. She called it "a certain irony" that "beyond human hunches AI itself is the most efficient tool we have for revealing what is AI generated."

Some people still use it to check what they share. On Wednesday, software developer Tim Bray said Pangram is "mostly ahead in the arms race against the AI vendors." He was writing about how he checks the links he shares.

Prizes Should Check Drafts, but That Asks a Lot of Writers

Prize officials say drafts and interviews are the fair test. Others say writers shouldn't have to hand over forensics to be believed.

Prizes should check drafts and talk to writers. Caine Prize chair Ellah Wakatama told Vara in October that anyone making such an accusation must offer proof that "does not come from a machine." On Wednesday she said, "we feel the evidence is clear: The Blue Fighter was human written." Razmi Farook, director general of the Commonwealth Foundation, told CNN that "the fairest approach is to consider multiple sources of evidence."

People who lean on software get worse at judging. Author John Warner told CNN that automated detection will "de-skill people from discerning these things for themselves."

Making writers prove themselves costs something too. Wong said Wednesday that it isn't "fair or sustainable to expect authors and artists to produce digital forensics documenting their creative process on demand." CNN writer T.M. Brown wrote that asking writers to hand over drafts to get cleared raises "thorny social questions." The Caine Prize is also adding new rules. When 2027 submissions open Nov. 23, publishers will have to submit an AI declaration for every story, and shortlisted writers will have to make one too.

Where This Lands

Wong says The Atlantic's claim never had evidence, and de Bodard says Pangram returned a false positive. Attiah and Calhoun say claims like this hit Black writers hardest. Spero says detectors help magazines handle a flood of AI submissions, and Bray says Pangram is mostly ahead of the AI vendors. Wakatama and Farook say drafts and interviews should decide these cases. Wong says writers shouldn't have to produce forensics on demand. The Caine Prize's AI declarations start with 2027 submissions on Nov. 23.

Sources