AI Writing Scandals Mount, but Proving Them Remains Nearly Impossible
The Commonwealth Short Story Prize and British literary magazine Granta faced backlash this week after readers questioned whether the winning entry and two finalists were generated by artificial intelligence. The accused writers stayed silent, leaving the publications to investigate on their own.
Granta asked Anthropic's ChatGPT-competitor Claude whether the winning piece used AI. Claude concluded it was "almost certainly not produced unaided by a human" - a non-answer that satisfied no one.
The decision to rely on Claude sparked further criticism. A general-purpose chatbot is not designed to detect AI-generated writing. If purpose-built AI detectors fail regularly, Claude would perform worse.
This pattern has repeated across the industry. The New York Times, Hachette, BenBella Books, and Sports Illustrated have all dealt with similar accusations. Each time, the core problem remains: there is no reliable way to prove AI authorship.
The Detection Problem
Suspicious writing patterns - awkward phrasing, strange word choices, clunky sentences - can suggest AI involvement. They are not proof. Plagiarism detectors can compare text against known sources. AI detectors have no equivalent baseline.
Nota News, a network of local news sites, published articles that appeared obviously AI-generated. Investigation revealed the real problem: the sites had copied reporting and quotes exclusive to other outlets. Only when the editors later admitted to using AI was the question settled.
Without technical proof, accused writers have little reason to respond. Established journalists working for major publications face reputational damage that forces them to speak. Independent writers and those with minimal online presence lack that incentive.
Silence as Strategy
As detection methods grow less reliable, writers may realize they can stay quiet and face minimal consequences. Revoking a prize or removing published work without proof raises legal and ethical questions that publications cannot easily answer.
One proposed solution is public shaming - the social cost of suspicion itself. Whether a writer used AI or simply absorbed its patterns enough to imitate them, reputational damage could serve as enforcement where technical detection fails.
For writers navigating this moment, the lesson is clear: the tools to catch AI-generated text do not yet exist. The burden of proof has shifted from detection to behavior, and behavior leaves room for plausible deniability.
Your membership also unlocks:
