The Caine Prize for African Writing is reviewing an allegation that its newly selected winning story was produced with generative AI, placing a literary award in the difficult position of evaluating both authorship and the reliability of software designed to detect machine-written text. The Guardian reported on October 14 that Zimbabwean writer Simbarashe Kundizeza has firmly denied using AI to write The Blue Fighter.

The Blue Fighter was announced as the prize winner on October 6. Five days later, author and journalist Vauhini Vara published an article in The Atlantic arguing that the story appeared to be AI-generated. Vara said she submitted the text to Pangram, an AI-based detection service, which classified nearly all of the story as artificially generated except for an epigraph attributed to South African writer Vangile Gantsho.

Vara also based her argument on literary judgment. She identified repeated sentence patterns, heavy use of metaphor and broad aphorisms as characteristics she associates with machine-produced writing. She further compared the winning story with a 2021 work by Kundizeza and said their tones and styles differed. Those observations remain interpretations of the writing rather than direct evidence about how the newer story was created.

Varied manuscript pages pass through an AI detector producing an ambiguous signal.
The allegation combines a detector result with subjective judgments about style and changes between works.

Kundizeza said in a public statement that he wrote The Blue Fighter himself and did not use generative AI. He also rejected the idea that a difference between two works written years apart should be treated as suspicious, arguing that writers can change genres, adopt new styles and improve their craft over time. The Guardian reported no admission or documentary evidence showing that AI was used.

The Caine Prize said it is investigating the allegation and intends to use a human-centered process that supports the writer while the review proceeds. Prize chair Ellah Wakatama challenged the idea that an AI detector should act as the final authority in the case. She also objected to publicly treating a winner as a cheat before a proper investigation and raised concern about black writers becoming test cases for immature detection technology.

The dispute highlights an asymmetry in AI-authorship claims. A detector can generate a confident classification quickly, but the person accused may be asked to prove a negative about a private creative process. Style comparisons can add context, yet they can also penalize experimentation or growth. The Guardian’s report does not say what evidence the prize will examine or when its review will conclude.

An author’s desk faces a balanced review table prepared for a careful investigation.
The Caine Prize says it will use a human-centered review rather than treating software as the final authority.

Vara cited research from Pangram claiming that its false-positive rate is low, including when evaluating English written by non-native speakers. She also referenced a peer-reviewed study that found strong accuracy when Pangram assessed mock academic papers written fully, partly or not at all with AI. However, she acknowledged in a BBC interview that relatively little independent research has tested the broader accuracy of AI-detection tools.

The case is not isolated. The Guardian noted that Trinidadian writer Jamir Nazir faced social-media claims of AI use after winning the Commonwealth Foundation’s short story prize in July. In another case, Canadian-Haitian author Thélyson Orélien’s novel was removed from the longlist for France’s Goncourt prize after an anonymous account alleged that it had been written almost entirely by AI.

For literary institutions, the immediate challenge is procedural. Awards may need standards that distinguish a detector’s output from verified evidence, specify what authors can provide in response, and prevent an allegation from becoming a verdict before review. The Caine Prize investigation may clarify its approach, but the current reporting leaves the central question unresolved: Kundizeza denies using AI, while the detector-based claim remains an allegation under examination.