Hachette pulls novel over suspected AI

Hachette Book Group has canceled the planned U.S. release of *Shy Girl*, a horror novel by Mia Ballard, and its U.K. publisher has stopped continuing that edition after the company concluded the book showed signs of generative AI use. The novel had an unusual path. Ballard first self-published it in February 2025. Hachette’s Wildfire imprint then released it in the U.K. in November 2025, and Orbit was set to publish it in the U.S. in April 2026 before the release was scrapped (thebookseller.com, parade.com). The striking part is not just that the book was pulled. It is how far it got first. Readers had been complaining for months that the prose felt machine-made. Reviews on Goodreads pointed to repetitive phrasing, odd formatting, and sentences that seemed to circle the same ideas again and again. A long YouTube breakdown pushed those suspicions wider. Then the issue jumped from reader gossip to publisher crisis after evidence was gathered for *The New York Times* and shown to Hachette (slate.com, thebookseller.com, independent.co.uk). That is what makes this episode matter. Hachette did not say the internet had won an argument. It said it had reviewed the manuscript and decided to act. The company told trade outlets that it remains committed to protecting “original creative expression and storytelling,” and confirmed that Orbit would not publish the book in the U.S. while Wildfire would stop publishing it in the U.K. The book was also removed from Hachette’s site and from Amazon listings tied to the publisher (thebookseller.com, independent.co.uk). Ballard has denied using AI to write the novel. According to follow-up reporting, she told *The New York Times* that someone she knew had edited the self-published version and may have introduced AI in that process. She has also said she is pursuing legal action and that the controversy has damaged her reputation. That defense does not really solve Hachette’s problem. If AI entered the text at the editing stage and no one caught it, the result is the same: a major publisher nearly launched a book whose authorship it could not confidently vouch for (techcrunch.com, parade.com, slate.com). That failure lands at a bad moment for publishing. The London Book Fair this March became a stage for a broader fight over AI, copyright, and disclosure. The U.K. Society of Authors launched a “Human Authored” scheme there on March 11, 2026, explicitly pitching it as a way to identify books written by people in a market “increasingly flooded” by AI-generated titles. The Authors Guild is backing a parallel certification effort in the U.S. The Publishers Association has also argued that fully AI-generated works should be labeled so readers know what they are buying (societyofauthors.org, authorsguild.org, thebookseller.com). The deeper problem is that none of these tools fixes the weak point that *Shy Girl* exposed. Detection is messy. Contract language is often vague. Editorial workflows were built to judge quality, not provenance. So a book can move from self-publishing to a major house, pick up blurbs and marketing copy about a writer’s “visceral” voice, sell nearly 1,900 tracked print copies in the U.K., and only then trigger a full authorship crisis (thebookseller.com, thebookseller.com).