Hachette Book Group cancelled the US release of horror novel Shy Girl by Mia Ballard last Thursday and pulled the UK edition from sale, after months of reader speculation and a New York Times investigation concluded the book was largely AI-generated. It is believed to be the first time a Big Five publisher has withdrawn a title over AI use.
Shy Girl was self-published in February 2025 and picked up by Hachette later that year. The UK edition went on sale in November 2025 and sold 1,800 print copies before being pulled. AI detection firm Pangram tested the text and estimated 78% of the book was AI-generated.
What went down
Reddit users and a book editor first flagged LLM-like prose patterns in January.
A YouTube video titled "i'm pretty sure this book is ai slop" hit 1.2 million views.
Hachette cancelled the US release and discontinued the UK edition on 19 March.
Author Mia Ballard denied using AI, blaming an editor she hired for the self-published version.
Ballard had previously admitted to using stolen cover art from artist Whyn Lewis.
Why does this matter?
Publishers are increasingly acquiring self-published titles that have already proven commercial traction. When a book arrives with sales figures attached, the editorial review is often minimal. Shy Girl appears to have passed through Hachette's acquisition process with very little scrutiny of the actual text.
Major publishing contracts now prohibit AI-generated content, but the ability to detect is becoming increasingly difficult. Readers on Reddit and YouTube identified the problem months before the publisher acted, and Hachette only moved after the New York Times made contact.
Our take
This made news because its the first incident of its kind with a major publisher. But it is not going to be the last - in the next few years it’s likely to become a tidal wave.
Book publishers have spent decades positioning themselves as gatekeepers of quality. Agents read manuscripts, editors shape them and copy editors clean them up. That process exists to catch problems before they reach a reader but Shy Girl appeared to skip most of it.
It was self-published, went viral, got acquired on the strength of its sales numbers, and published in the UK with what industry observers describe as minimal editing. The AI detection was effectively crowdsourced by readers who were better at spotting the patterns than the professionals paid to evaluate the book.
Writer Lincoln Michel made the observation that when a publisher acquires an already-published title, extensive editing is rare. As publishers lean harder on self-published and BookTok-driven hits to fill their catalogues, the gap between what gets acquired and what gets properly vetted will widen.
But a key point is this is AI as it is now. In a year or so, AI writing may no longer be so obvious and many of its common traits could be removed. So essentially, AI writing will become indistinguishable from human writing, and publishers simply won’t be able to tell.
Another big thing... NVIDIA's $1 trillion inference bet
Something a bit more technical. NVIDIA used its GTC 2026 keynote last Monday to unveil the Vera Rubin platform and the Groq 3 LPU, its first non-GPU inference chip, built from the $20 billion Groq acquisition completed on Christmas Eve. CEO Jensen Huang projected $1 trillion in purchase orders through 2027 and proposed giving engineers annual AI token budgets worth roughly half their base salary.
The strategic signal is that GPUs alone cannot serve the agentic workloads NVIDIA is selling. Rubin handles training and prefill while Groq chips handle low-latency token generation, a split that delivers 35x more tokens per watt according to the company. NVIDIA is building the infrastructure for inference-first computing, and it spent $20 billion to make sure nobody else got there first.
