Recently, the literary agent Kate Nash started noticing that the submission letters she was receiving from authors were becoming more thorough \u2013 albeit also more formulaic.<\/p>\n
\u201cI took it as a rise in diligence,\u201d she said. \u201cI thought it was a good thing.\u201d<\/p>\n
But then she had what she described as her eureka moment: the letter with the AI prompt right at the top. \u201cIt read: \u2018Rewrite my query letter for Kate Nash including a comp to a writer she represents,\u2019\u201d she said.<\/p>\n
Once Nash had seen the prompt, she \u201ccouldn\u2019t unsee AI-assisted or AI-written queries again\u201d.<\/p>\n
Hachette\u2019s Orbit imprint confirmed it had halted US publication of Shy Girl, by Mia Ballard, after an internal review. The title has also been removed from online retailers and will no longer be distributed in the UK. Photograph: PR<\/p>\n
The news last week that Mia Ballard\u2019s \u201cfemgore\u201d horror novel Shy Girl could be up to 78% AI-generated<\/a>, however, has forced literary agents and publishers alike to consider whether sharp eyes alone can detect AI-generated work.<\/p>\n \u201cThe question of how Shy Girl slipped through Hachette\u2019s net is something the publisher has to answer themselves, but in reality, it was only a matter of time before this happened,\u201d said Anna Ganley, the chief executive of the Society of Authors.<\/p>\n Wildfire, a UK imprint of Hachette, had published Shy Girl in November 2025. It was due for US publication in April, but the controversy led to its UK discontinuation and US cancellation earlier this month. Ballard has denied using AI to write Shy Girl, telling the New York Times<\/a>, which first reported the story, that an acquaintance she hired to edit a self-published version of the novel had used it.<\/p>\n An editor at one of the \u201cbig five\u201d publishing houses said a \u201ccold shiver went down my spine\u201d when the Shy Girl story broke. \u201cIt really is a case of \u2018there but for the grace of God go I,\u2019\u201d they said.<\/p>\n \u201cIt\u2019s an issue publishers are keenly aware of. We make it very clear to authors what we expect, we get them to sign contracts and we run their work through multiple AI detection tools, but we know all this is fallible.<\/p>\n \u201cHence the cold shiver: if an author is determined to use AI, then cover their tracks, there\u2019s very little we can do.\u201d<\/p>\n Prof Patrick Juola, a US computer scientist known for his work on authorship attribution, agreed. \u201cI don\u2019t want to call AI detection tools a scam, but it\u2019s a technology that simply doesn\u2019t work.\u201d<\/p>\n He likened the failure to antibiotic resistance: \u201cAI is a learning system continually upgraded by its manufacturers. If there was a detection technology that worked, then people would simply build better AI tools to fool it.\u201d<\/p>\n Mor Naaman, a professor of information science at Cornell Tech and head of its social technologies research group, agreed. \u201cAI learns very quickly how to avoid AI detection. We\u2019re not quite there yet, but soon publishers won\u2019t stand a chance,\u201d he said.<\/p>\n Already, the sophistication of the technology threw up an interesting point, said Nikhil Garg, an assistant professor at Cornell Tech\u2019s Jacobs Institute. \u201cSophisticated authors who want to evade the detection tools know how to edit their text, test it against these tools and revise again,\u201d he said. \u201cAt some point, you have to ask: has it become their own work anyway, despite the AI?\u201d<\/p>\n Naaman agreed that while Shy Girl appeared to be an \u201cegregious\u201d example, there were increasingly grey areas. \u201cWe all work in an AI-hybrid world now. When does something become an AI-generated book, rather than just using AI like I use a spellchecker, to fix my grammar or maybe spark ideas?\u201d he asked.<\/p>\n