Hachette, Elsevier, Cengage Learning and author Scott Turow are suing Google, claiming it trained Gemini on millions of books after promising publishers it would only show snippets. If the court agrees, Google's book problem gets much bigger than a fair use debate.
Hachette Book Group, Cengage Learning and Elsevier walked into federal court in Manhattan on July 13 and accused Google of doing exactly what it warned its own employees not to do. The publishers, joined by bestselling author Scott Turow, say Google trained its Gemini models on millions of their books without paying for them or asking first. That's the charge.
A Promise About Snippets, Broken
The lawsuit, filed in the U.S. District Court for the Southern District of New York, centers on a specific promise. Publishers submitted books to Google Books, Google Play and Google Scholar under agreements that let Google display short snippets to readers searching for a title. That's the whole deal. Nothing in those terms authorized Google to feed the full text into a large language model built to compete with the books themselves. The plaintiffs call it a bait and switch: a search feature, repurposed into training data.
The complaint leans hard on Google's own paper trail. According to reporting from The Guardian, the plaintiffs allege that internal Google documents flagged book training as "highly problematic" and estimated possible copyright exposure in the range of "$10Bs-$100Bs." Google proceeded anyway, the suit says, and in some cases stripped or altered copyright management information to obscure where the training data came from.
That's not a rounding error. It's a company doing the math and deciding the lawsuit might be cheaper than the license.
The plaintiffs want statutory damages, a permanent injunction stopping further use of the material, and an order forcing Google to destroy any unauthorized copies it holds. An aggressive list of demands. Publishers Weekly reported the suit is structured as a putative class action, which means it could eventually cover a far wider pool of rightsholders than the four named plaintiffs.
The Fair Use Fight Gets Messier
Google isn't the first AI company to get sued over books, and it won't be the last. Anthropic agreed to pay $1.5 billion to settle a similar claim from authors including Andrea Bartz, Charles Graeber and Kirk Wallace Johnson, in what's been described as the largest copyright recovery in U.S. history. Judge William Alsup gave that settlement preliminary approval in September 2025. Under its terms, Anthropic pays roughly $3,000 for each of about 465,000 books, according to the Associated Press.
Alsup's earlier ruling in the Anthropic case drew a sharp line. Training an AI on books the company bought and scanned could qualify as fair use. Keeping a central library built from pirated copies could not. Google's publishers are testing a third scenario: books that were neither pirated nor licensed for AI, but handed over under contracts that talked about search, snippets and distribution, not Gemini.
That's the real exposure here.
If you're a publisher, this case is not only about old books sitting in a database. It's about whether a platform can take material supplied for one narrow product and quietly make it fuel for another, far more valuable one. That's why the snippet promise matters. Contract scope is dry until a company uses it to build a model that can summarize, imitate and compete with the work that made the model useful in the first place.
Google hasn't filed a response yet. Reporters at TechCrunch and Publishing Perspectives both noted the company had no comment on the specific allegations as of this week. Google is already fighting similar book-training suits alongside OpenAI and Meta in courts around the country. Here, the scale is different. Three of the largest publishers in the country, plus Turow, are now saying Google knew the price tag and signed up for the fight anyway.
Frankly, that is the harder case for Silicon Valley to wave away. A random scraped web page is one argument. A book delivered to Google under a specific commercial arrangement is another. The court now has to decide whether permission to show a few lines to a reader also means permission to ingest the whole work into Gemini.
Also read: IBM Suffered Its Worst Trading Day Since 1987 As Clients Chased AI Hardware • Adam Mosseri Says AI Token Costs Could Soon Match Engineer Salaries • Meta Is Now Rationing AI Tokens Like It Rations Headcount
