Three of the world's larger book publishers, along with the bestselling novelist Scott Turow, have taken Google to court over how it built its AI. Hachette, Cengage and Elsevier filed a class-action lawsuit in the US District Court for the Southern District of New York on July 14, alleging that Google trained its Gemini AI models on their copyrighted books without authorization, according to TechCrunch. The claims are allegations that a court has not tested, and Google has not publicly responded.
What the publishers allege
At the core of the complaint is a claim that Google used the plaintiffs' books to train Gemini without permission, and then tried to hide it. The suit alleges that Google "intentionally removed or changed copyright information" so that the models "were trained on stolen materials," per TechCrunch's account of the filing. The publishers say Google obtained many of the books through limited-purpose programs such as Google Books and the Google Play store, then reused them for a different and commercial purpose, building an AI system, beyond what those programs allowed.
The filing also points to Google's own awareness of the risk. It cites an internal company document warning that using copyrighted books for AI training could be "highly problematic for Google" and might lead to "$10Bs-$100Bs in potential fines," according to TechCrunch. The plaintiffs, who are seeking to represent a class of rights holders, want damages and a court order to stop the alleged infringement.
Part of a bigger fight
The case does not stand alone. Publishers, authors, news organizations and music companies have filed a string of copyright suits against AI developers over the past two years, arguing that training models on their work without a license is theft dressed up as innovation. AI companies have generally countered that training is a transformative "fair use" of publicly available material. Courts have so far sent mixed signals, and at least one prominent AI developer has chosen to settle with authors rather than fight to a verdict.
Because it was filed in New York rather than California, where some earlier fair-use rulings favored AI developers, this suit could produce a different reading of the law. Different courts are not bound by one another's conclusions, which is part of why the outcome of these cases remains genuinely uncertain.
Why the money matters
Underneath the legal argument is a business one. Publishers say AI tools that answer questions directly, using knowledge distilled from their content, both take their material for training and then divert readers who might otherwise visit their sites or buy their books. If courts agree that training on copyrighted work requires payment, AI developers could face large damages and, more lastingly, the need to license content, turning today's free training data into a recurring cost. If courts side with the AI firms on fair use, publishers lose a key source of leverage over how their work is used.
For now, the specifics are contested and unproven, and Google has yet to make its case in public. What is clear is the size of the stakes: the same internal figure the plaintiffs cite, potential fines in the tens of billions, is a measure of how much rides on where the law lands. Boursel will report the court's findings rather than predict them.
