AI Detector Sparks Witch Hunt Over Caine Prize Winner
A major literary prize winner is under investigation after an AI detector flagged his story, proving we cannot trust machines to police art.
Drama · Source: Brittle Paper
What happened
Zimbabwean writer Simbarashe Steyn Kundizeza won the 10,000 pound Caine Prize for African Writing for his short story The Blue Fighter. Days later, The Atlantic published an article suggesting he used artificial intelligence. Journalist Vauhini Vara ran the text through an AI detection tool called Pangram. The software flagged almost the entire story as machine-generated, clearing only the epigraph. Vara noted stylistic shifts from his 2021 work, pointing to specific sentence structures and metaphors as evidence of chatbot writing.
Kundizeza flatly denies the claim. He posted a detailed timeline on X showing his drafting process from May to June 2025. He cited specific emails with early reader Vangile Gantsho and his editor at The Shallow Tales Review. He pointed out the story is deeply personal. It is based on his father quitting his accounting job to grow tomatoes during Zimbabwe's 2002 economic collapse. He argued that writers naturally change their style over five years depending on the subject matter.
The Caine Prize board is now investigating the controversy. Chair Ellah Wakatama says the prize will review Kundizeza's drafts, interview the judges, and speak with the author about his methods. However, Wakatama publicly rejected Pangram as the ultimate authority. She stated the tool is not yet tested and demanded proof beyond a machine's output. The African literary community has rallied behind Kundizeza, with publishers calling the reliance on AI detectors wildly unethical.
Key facts
- 10,000 — The prize money in pounds awarded to the Caine Prize winner.
- 1 in 24,000 — The false positive rate Pangram claims for its AI detector.
- 25 — The number of previous Caine-winning stories Pangram rated as human-written.
Why it matters
AI detectors are becoming weaponized against creators. Builders are shipping tools like Pangram to publishers, schools, and employers with massive claims of accuracy. The Atlantic cited Pangram's false positive rate as just one in 24,000 cases. But when these tools inevitably flag human work, they create massive reputational damage. A machine spits out a probability score, and a human writer is suddenly forced to defend their entire career. Resolving these false positives takes weeks of manual review and public relations nightmares.
Trust in digital text is collapsing completely. As generative AI floods the internet, publishers are panicking. They are grasping for automated solutions to filter out the noise. This leads to a guilty-until-proven-innocent environment for anyone publishing text today. Writers now have to meticulously log their keystrokes, save every rough draft, and document their inspiration just to prove they are human. The burden of proof has shifted entirely onto the creator.
For builders
Build proof of work tools for writers
The market desperately needs tools that cryptographically verify the human writing process. Version control for prose is no longer just a feature for collaboration. It is now a required defense against AI accusations. Whoever builds the standard for verifying human keystrokes will own the publishing market.
Stop selling AI detectors as absolute truth
Selling AI detection with near-perfect accuracy claims is a massive liability. When your tool inevitably flags a human, your brand takes the hit. Position these tools as filters or warning systems, not final judges. You must educate your users on false positives.
Prepare for manual audit workflows
Platforms cannot rely solely on automated moderation for AI content. You must build escalation paths for human review. If you ban users or reject submissions based purely on a machine score, you will lose your best creators. Build the infrastructure for appeals.
My take
AI detectors are snake oil. You cannot reliably use a neural network to catch another neural network without destroying innocent people in the crossfire. If publishers want to verify human authorship, they need to look at the writer's process. Running a text scanner and publishing a hit piece is lazy journalism and terrible technical practice.
Original reporting: Brittle Paper. This is my rewrite and opinion.