Featured image: A man with a shaved head and a weathered face is seated at a wooden table, writing in an open notebook with a pen. AI-generated image; not a photograph of the reported event.
Zimbabwean author Simbarashe Kundizeza has denied using artificial intelligence to write his award-winning short story, The Blue Fighter, which won this year’s Caine Prize for African Writing. Source: bbc.co.uk.
What Happened
The denial comes after an article in The Atlantic claimed the story was entirely generated by AI, based on analysis from the AI detection tool Pangram. The article, titled Yet Another Prizewinner Seems to Be AI, was published on Sunday and sparked immediate backlash across social media platforms, with many users expressing concern over the implications for authorship, creativity, and equity in literary recognition.
Kundizeza responded on social media, stating he wrote the story and provided a clear chronology of its development. He emphasized that the work has a documented history of inspiration and personal connection to his life. In a post on X, he wrote: For the avoidance of doubt, I wrote The Blue Fighter.
The post received widespread support from fellow writers, readers, and literary communities across Africa and beyond.
The story was first published by The Shallow Tales, a literary journal known for its focus on African voices and experimental storytelling. Kundizeza described the inspiration behind the piece as stemming from a poem he heard at an event in Ghana, where South African poet Vangile Gantsho performed a piece about her father’s obsession with building fish tanks. He noted that the imagery of a man devoted to constructing intricate, enclosed systems resonated with his own family’s experiences during Zimbabwe’s economic downturn in the early 2000s.
He explained that his own father, an accountant, had quit his job to grow tomatoes in a small backyard plot. This personal anecdote, he said, formed the emotional core of the story. ‘Much like the father in the story, my own dad… was an accountant, but he quit his job to grow tomatoes. So it kind of touches on both of our fathers,’ he told BBC Newsday. This connection allowed him to blend fictional elements with real-life experiences, creating a narrative that reflects both personal memory and broader socioeconomic realities in Zimbabwe.
Key Facts
- The story was first published by The Shallow Tales.
- Kundizeza says he was inspired by a poem read at a Ghana event about a man obsessed with building fish tanks.
- He linked that inspiration to his own father, who quit his job as an accountant to grow tomatoes during Zimbabwe’s economic downturn in the early 00s.
- The Caine Prize chair, Ellah Wakatama, said she was shocked that The Atlantic published its claim without a full investigation.
- Wakatama stated that AI detection tools like Pangram are not yet proven reliable, especially for non-native English writing.
- The Caine Prize does not conduct AI audits of submissions. Writers are expected to vouch for originality, and no formal checks are performed on entries for AI content.
- Vara’s article cited research from the Free University of Brussels, which found that AI detection tools like Pangram incorrectly flagged zero cases of English-language writing by non-native speakers as AI-generated.
- Vara personally tested earlier Caine Prize-winning stories and found they all came back as human, including a story by Kundizeza that had a different tone from a later piece.
- Kundizeza responded that the idea that writers cannot improve or shift in style over time is a fundamental misunderstanding of the craft of writing.
Background: How AI Detection Tools Work
AI detection tools like Pangram operate by scanning written text for linguistic patterns commonly associated with machine-generated content. These include repetitive sentence structures, overuse of specific phrases, and the frequent use of metaphors or emotional expressions that follow predictable templates.
One such pattern cited in the article is the use of constructions like X but Y
, which Vara said is common in AI-generated writing. For example, she pointed to a passage describing a mother’s emotional state: Not with anger, but with the quiet resignation of someone who had known all along this moment would come.
She noted that such phrasing, while emotionally resonant, appears frequently in outputs generated by AI systems trained on large corpora of human text.
However, the effectiveness of these tools is limited by their training data and design. AI detectors are typically trained on vast databases of human-written content, but they may not account for cultural nuances, regional dialects, or the stylistic evolution of individual writers. As a result, they often struggle to distinguish between human creativity and machine-generated output when the writing is emotionally complex or stylistically varied.
Research conducted by the Free University of Brussels found that AI detection tools like Pangram incorrectly identified zero cases of English-language writing by non-native speakers as AI-generated. The study concluded that such tools are not yet capable of reliably distinguishing between human and machine authorship in diverse linguistic contexts. The researchers recommended that AI detection results be treated as preliminary indicators, not definitive judgments, especially in literary analysis where style, voice, and personal experience are central.
Vara, who authored the The Atlantic piece, acknowledged that her own testing of past Caine Prize-winning stories showed consistent results: all came back as human. She observed that a previous story by Kundizeza had a different tone from a later piece, yet both were authored by the same writer. This observation led her to conclude that the stylistic evolution of a writer over time is a legitimate and expected phenomenon, not a sign of AI involvement.
Kundizeza responded to this point by emphasizing the natural progression of a writer’s craft. He stated that the idea that a writer cannot develop or shift in style over time is a ‘fundamental misunderstanding of the craft of writing.’ He argued that growth in storytelling comes from experience, reflection, and exposure to diverse forms of expression—processes that are inherently human and not replicable by algorithms.
Why It Matters
The controversy surrounding Kundizeza’s story highlights a broader and growing tension in the literary world: the increasing use of AI detection tools to evaluate human creativity. While such tools may offer efficiency in content audits, their application in high-stakes literary awards raises serious ethical and practical concerns.
Ellah Wakatama, chair of the Caine Prize and editor-at-large at Canongate Books, has publicly expressed concern that the use of AI tools like Pangram as final arbiters in authorship decisions undermines the integrity of literary awards. She stated: I am not happy for the work of writers, and in this case especially black writers, to form the testing ground for something that is at best still rough and ready.
Her comments reflect a wider apprehension among literary professionals that AI tools, still in developmental stages, may be used to unfairly scrutinize writers from marginalized backgrounds.
Wakatama emphasized that the Caine Prize does not check submissions for AI content. Instead, the process relies on the author’s personal vouching for originality. This approach is rooted in the principle that literary creation is a deeply personal, embodied, and cultural practice—one that cannot be reduced to algorithmic patterns.
Similar tensions have emerged in other literary awards. The Commonwealth Short Story Prize was won by Trinidadian writer Jamir Nazir despite allegations that he used AI. Canadian-Haitian author Thélyson Orélien withdrew from France’s top literary award, the Goncourt, after facing similar accusations, which he denies. These cases illustrate a pattern of increasing scrutiny of African and non-Western writers, raising questions about bias and the role of technology in shaping literary legitimacy.
Responses from the Literary Community
Several African writers and publishers have criticized the allegations as racially biased and ethically problematic.
Nigerian publisher Modupe Daramola called the accusations ‘incredibly sinister’ and questioned why an AI-powered tool like Pangram should be treated as definitive. She wrote on X: Pangram is itself AI-powered. Why should its assessment be treated as definitive?
Her argument underscores a paradox: using an AI tool to judge human creativity, while the tool itself is built on AI-generated training data and may be prone to error.
Nigerian writer Wale Lawal described The Atlantic‘s article as ‘crazy, racist and wildly unethical.’ He argued that the accusations disproportionately target writers from Africa and the Global South, where linguistic diversity and non-standard English usage are common. He stressed that such claims risk reinforcing existing power imbalances in the literary world.
Joe Ruzvidzo, a Zimbabwean novelist whose novel Welcome to Anywhere was published last year, stood by Kundizeza and said it was time to ‘fight this literary stop-and-frisk.’ He warned that treating writers as subjects of technological audits could erode trust in the creative process and discourage emerging voices from participating in literary discourse.
Limitations and Open Questions
AI detection tools are not foolproof. They rely on statistical patterns and may misidentify human writing, especially when authors use non-standard or regional language. For instance, many African writers use idioms, syntax, and metaphors that differ from standard English, making them more vulnerable to being flagged as AI-generated.
Research shows that AI tools perform poorly on English-language writing by non-native speakers. This raises concerns about fairness in literary evaluation, particularly when such tools are used to make final judgments about authorship.
There is no official standard for how AI tools should be used in judging literary works. The Caine Prize has not yet conducted a full investigation into the claim, though it is reviewing Kundizeza’s drafts and communications with judges. This lack of transparency has prompted calls for greater accountability and independent audits in future literary prize evaluations.
Wakatama said the Caine Prize does not check submissions for AI content. Writers are expected to vouch for originality, not submit to automated audits. This policy reflects a broader commitment to preserving the human element of literary creation—where voice, memory, and lived experience are central to the work.
Vara conceded that no AI detector is 100% accurate. She acknowledged that Pangram’s results are not infallible and should be treated as a starting point, not a conclusion. She also noted that her own experience with past Caine Prize stories reinforced the idea that AI tools are not reliable indicators of authorship in complex, emotionally rich narratives.
What to Watch Next
The Caine Prize is expected to release a formal statement on the matter in the coming days. It may include a full review of the story’s origin and authorship, as well as a public explanation of how the prize evaluates submissions.
Other literary prizes may face similar scrutiny. The debate could influence how future awards evaluate authorship and originality, potentially leading to new guidelines or policies on the use of AI tools in literary judgment.
Wakatama emphasized that the use of AI tools in literary evaluation remains experimental. She urged caution, especially when applied to works by writers from diverse cultural and linguistic backgrounds. As AI tools become more common in content creation, the line between human and machine-generated writing will remain a sensitive and evolving issue.
For more on African literature and writing, see our coverage on Zimbabwean literary achievements.
Further reading on AI and creativity is available in our recent reports on technology in literature.
Stay updated on African writing and global literary trends through our Africa news section.
