AI-writing scandals are becoming common. Proving them is another matter. - Poynter

Organizers of the Commonwealth Short Story Prize and Granta faced backlash after social media users questioned whether a winning entry and two finalists were generated by artificial intelligence. Similar scandals have appeared across major media organizations, but the Granta case drew attention because most accused writers did not respond. Granta said it asked Anthropic’s chatbot Claude whether the winning entry used AI, and Claude concluded it was almost certainly not produced unaided by a human. Further criticism followed because Claude is a general chatbot and not designed to detect AI writing. The situation may lead writers to stay silent, since proving AI use is difficult and revoking prizes without proof raises moral and legal concerns.

"Organizers of the Commonwealth Short Story Prize and British literary magazine Granta faced sharp backlash this week after social media users questioned whether the winning entry and two of the finalists were generated by artificial intelligence. That kind of scandal - writing that is suspected to have been generated with AI - has become increasingly common in recent months. The New York Times ( twice), Hachette, BenBella Books and Sports Illustrated have all suffered versions of it."

"In the wake of their silence, Granta released a statement saying that it had asked Anthropic's AI chatbot Claude whether the winning entry had used AI, and Claude determined that it was "almost certainly not produced unaided by a human." Given those inconclusive results, Granta has decided to keep the entries published on its website. The use of Claude to examine the entry provoked further backlash, according to Vara, because as a general chatbot, it is not designed to detect AI-generated writing."

""if purpose-built AI detectors can make mistakes in flagging AI prose, Claude could be expected to perform even more poorly." As a result, accused writers may decide to stay quiet in the future, knowing that it is very difficult to prove that they used AI, Vara writes. "Knowing that detection platforms are fallible-proving AI use isn't as simple as proving, say, plagiarism from another author's work-writers could be discovering an enforcement loophole ... revoking a prize without proof is, morally and legally, no simple matter.""

Read at Poynter