Gemini Trained with Millions of Copyrighted Books: Publishers and Authors Sue Google

Four major names in publishing and one of the most well-known authors of American legal fiction have taken Google to court. The lawsuit, filed by publishers Hachette, Cengage, and Elsevier along with writer Scott Turow, accuses the company of copying millions of copyrighted works to train Gemini without ever requesting permission or paying royalties.

The legal document speaks plainly of "one of the most prolific violations of copyrighted material in history." The publishers argue that Google reused texts originally provided for well-defined services, such as Google Books, Google Play Books, and Google Scholar, where agreements only allowed specific uses: showing extracts in searches or selling ebooks, not copying content to train a commercial AI product.

The lawsuit also cites internal discussions at Google in which the company allegedly acknowledged the legal risks of the operation, estimating a potential exposure of between $10 and $100 billion in potential penalties for the use of texts obtained through Google Play Books agreements. An internal calculation, according to the publishers, that did not stop the company.

The Accusation: An At-Risk Publishing Market from Generative AI

The crux of the lawsuit not only pertains to the act of copying but to the consequences on the book market. Publishers maintain that AI-generated content could directly replace the original works on which Gemini was trained, harming sales for authors and publishing houses.

The example cited in the lawsuit text is direct: Gemini could generate "a 100-page mystery set in a quiet seaside town full of secrets, replacing an original copyrighted detective novel that the model was trained on" in just 20 minutes and at a cost of 39 cents. "No publisher or author can compete with this," the document states.

Among the works that the publishers indicate were used without permission are well-known titles like The Fifth Season by N.K. Jemisin and Who Could That Be at This Hour? by Lemony Snicket. The list reportedly includes several other texts covered by limited licensing agreements with Google.

The authors of the lawsuit aim not only to contest the collection of training data but, above all, to highlight the fact that those books were granted to Google with precise contractual constraints intended for editorial uses and not to feed a language model designed to generate competing content. This is the gap, according to the publishers, between lawful use and what they term a "betrayal of the original agreements."

This is just the latest lawsuit already filed against major generative AI companies for the use of protected material for training models. However, the Google case adds a specific element: it is not a generic use of content retrieved online, but the alleged violation of already signed licensing contracts that were circumscribed for purposes other than AI training.