Publishers sue Google over Gemini AI training

Alphabet’s Google is facing a new class action over how it built Gemini, its artificial intelligence platform. For investors watching the AI race, the case adds another legal pressure point around a key question: whether tech companies can use copyrighted books to train models without paying rights holders.

The complaint, filed in the U.S. District Court for the Southern District of New York, was brought by publishers and authors including Hachette, Cengage, Elsevier, author Scott Turow and S.C.R.I.B.E. A class action is a lawsuit that seeks to represent a wider group of people or businesses with similar claims.

The plaintiffs allege that Google used copyrighted works to train Gemini without getting the permissions required by law, according to the complaint. AI training means feeding large volumes of text, images or other material into a model so it can learn patterns and generate new responses.

The complaint also accuses Google of removing or altering copyright information tied to the works. The plaintiffs allege that this was done to hide that Gemini models had been trained on material they describe as stolen, according to the lawsuit.

Why Google Books is part of the fight

The publishers’ case leans on their long relationship with Google. According to the complaint, authors and publishers previously gave Google access to copyrighted books for limited programs such as Google Books, where users can search for books and see short snippets, rather than read full copies.

The plaintiffs claim Google used works from those limited-purpose programs to train Gemini, as well as books uploaded to the Google Play store, without authorization. In the complaint, the plaintiffs say Google copied works from programs with restricted uses for AI training despite knowing it did not have permission.

The lawsuit also cites an alleged internal Google document that described using copyrighted books for AI training as “highly problematic for Google” and warned of “$10Bs-$100Bs in potential fines,” according to the complaint.

Google did not immediately respond to TechCrunch’s request for comment.

A broader fight over AI data

The case joins a growing stack of copyright disputes against major AI companies, including Google, Meta, OpenAI and Anthropic. Publishers, authors and other rights holders have argued that AI companies built valuable systems by ingesting protected material without licenses.

AI companies have often pointed to fair use, a U.S. copyright doctrine that can allow limited use of protected works without permission in certain circumstances. Two early California rulings cited by TechCrunch favored AI companies, finding that training on copyrighted books can qualify as fair use under U.S. copyright law.

Those rulings do not decide the Google case. The new lawsuit is in New York federal court, where a different judge can assess the claims and Google’s expected defenses.

The legal picture also remains mixed. TechCrunch reported that Anthropic was fined $1.5 billion for pirating works used in training, described as the largest payout in U.S. copyright law history. About half a million writers were eligible for at least $3,000 each, though many authors opted out so they could pursue additional claims over AI training, according to the Authors Guild.

For Alphabet shareholders, the lawsuit does not answer whether Google will owe damages or need to change its AI practices. It does show that the cost of AI development is no longer only about chips, cloud infrastructure and talent. The data used to train models is becoming a legal and financial battleground of its own.

This story draws on original reporting from TechCrunch.