Pangram flags four of ten Commonwealth Prize winners as AI

Four of the ten 2025 and 2026 regional winners of the Commonwealth Short Story Prize have now been flagged as AI-generated by Pangram, according to reporting in the Financial Times and a longer walk-through in The Argument.

The case most people are arguing about is 'The Serpent in the Grove', the Caribbean regional winner by Trinidadian writer Jamir Nazir, which Pangram scored as 100% machine-written after its publication in Granta. Nazir denies using AI at all; he told The Observer his writing process runs through the text-to-speech function on his Android phone. Pangram Labs then went back through the archive. 'We tested every Commonwealth Short Story Prize winner since 2012,' the company posted, saying it had found two more flagged stories among the 2026 regional winners ('The Bastion's Shadow' from Canada and Europe, 'Mehendi Nights' from Asia) plus the 2025 overall winner.

Granta publisher Sigrid Rausing has been unusually blunt about the uncertainty. 'It may be that the judges have now awarded a prize to an instance of AI plagiarism, we don't yet know,' she said. The Commonwealth Foundation, which received 7,806 entries this cycle and awards £2,500 to each regional winner and £5,000 to the overall, is reviewing its selection process but has not withdrawn any prizes, saying that 'all shortlisted writers have personally stated that no AI was used' and that until reliable detection exists it 'must operate on the principle of trust'.

Pangram is also the detector behind the biggest publishing casualty of the year. Hachette Book Group pulled Mia Ballard's horror novel 'Shy Girl' from shelves this spring after CEO Max Spero said the book was '78 percent AI generated'. Ballard denies writing with AI and says a hired editor used it; she is Black and has said writers of color are being disproportionately scrutinised. Nigerian author Jerry Falade's crime novel 'Call Me, I'll Hide the Body', reportedly acquired for a multimillion-dollar deal and slated for 2028, was withdrawn by his agency after similar questions.

Pangram's defence rests on a 2025 University of Chicago study that reported 'a 0.005 or lower false positive rate' for the tool. What the flagged authors describe is what would happen at the other tail: polished human prose landing at or near 100. Anthropic itself recently documented gaps in its own Claude watermark on code and rewrites, a reminder of how much these tools still miss at the edges. Our AI-detection feed has logged thirty other stories in the last three months; this is the one where a detector's verdict has already ended a novelist's book deal.