Granta uses Anthropic's Claude to detect AI-written entry, but fails to prove it

The latest AI writing scandal erupted when social media users questioned whether the winning entry and two finalists of the Commonwealth Short Story Prize, also published by Granta, were AI-generated. Unlike previous scandals involving The New York Times, Hachette, or Sports Illustrated—where accused writers typically spoke out—most accused authors in this case remained silent. In response, Granta turned to Anthropic's chatbot Claude, asking it to analyze the winning entry. Claude returned an inconclusive assessment, stating the text was "almost certainly not produced unaided by a human." The move sparked further backlash, as critics argued that Claude is a general-purpose chatbot, not a dedicated AI detector, and is even less reliable than purpose-built tools, which themselves make frequent mistakes.

This incident underscores the broader challenge of proving AI-generated content. As Atlantic contributor Vauhini Vara noted, detection platforms are fallible, and proving AI use is not as straightforward as proving plagiarism. Accused writers may exploit this enforcement loophole by staying quiet, knowing that revoking a prize without solid proof is morally and legally fraught. Independent journalists and less-established authors, lacking the reputation incentives of legacy writers, are especially likely to remain silent. The growing sophistication of AI-generated text means publishers and prize committees must rethink how to verify authenticity in an era where the line between human and machine writing continues to blur.