Simbarashe Kundizeza won the 2026 Caine Prize for African Writing on 6 October with his short story "The Blue Fighter." Five days later, the Atlantic's Vauhini Vara published an article declaring the story "seems to be AI," citing Pangram, an AI-detection tool that flagged the text as almost entirely machine-generated. Kundizeza denied the allegation categorically. The Caine Prize launched a review. The evidentiary chain here is alarmingly thin. Vara ran the story through one commercial detector, noted stylistic features she associated with AI — "not x, but y" constructions, heavy metaphor, aphoristic phrasing — and compared it unfavorably to a 2021 story by the same author. Kundizeza's response was blunt: suggesting a writer cannot evolve over five years "fundamentally misunderstands the craft of writing." The Caine Prize chair, Ellah Wakatama, was sharper still, objecting that black writers were being used as "the testing ground for something that is at best still rough and ready." Pangram's own research claims low false-positive rates, and one peer-reviewed study found it had "strong accuracy" on mock academic papers. But mock academic papers are not literary fiction. Vara herself acknowledged "relatively little third-party research" on detector accuracy. That concession sits awkwardly next to the certainty of the headline. A tool with limited independent validation is being treated as a forensic instrument in a domain it was not validated for. This is not an isolated incident. Jamir Nazir's Commonwealth Foundation prize win in July drew similar social-media allegations. Thélyson Orélien's novel was pulled from the Goncourt longlist in September after anonymous AI claims. A pattern is forming: literary prizes attract AI accusations, detection tools supply a veneer of objectivity, and the accused writer — often from the Global South — bears the reputational damage regardless of outcome. The structural incentive problem is clear. AI-detection companies benefit from high-profile cases that demonstrate their relevance. Journalists benefit from the outrage cycle. Prize committees, caught between credibility and duty of care, are forced into reactive investigations. The writer, meanwhile, faces a burden of proving a negative — that they did not use a tool — with no established forensic standard for what constitutes proof. Vara's stylistic critique — that the story is "bad" and "maudlin" — is doing separate work from the AI-detection claim, but the two are presented as mutually reinforcing. This conflation is dangerous. Bad writing is not evidence of AI generation. Plenty of human writers produce heavy metaphor and tired tropes. Plenty of AI-generated text reads cleanly. Aesthetic judgment and forensic detection are different disciplines, and blending them erodes the credibility of both. The Caine Prize's stated commitment to a "human-centred approach" is the right instinct, but the institution is navigating without a map. No literary prize has established protocols for AI-authorship disputes. No independent body certifies detection tools for creative writing. Until those structures exist, every accusation is an ad hoc trial where the tools are unvalidated, the standards are undefined, and the accused has no clear path to exoneration.
