Google is facing another copyright lawsuit over books and journal articles allegedly used to train Gemini.
Three major publishers and author Scott Turow filed a proposed class-action lawsuit alleging Google copied copyrighted books and journal articles to develop its Gemini AI models, according to the Association of American Publishers.
A key issue is whether files supplied for search tools and ebook sales could later be used for model training.
Internal warnings detail the legal risk
Internal discussions cited in the complaint indicate employees had considered possible consequences of using publisher-provided books.
According to the complaint, one internal assessment described the practice as “highly problematic for Google.” Another warned of “$10Bs-$100Bs in potential fines.”
Gemini’s lead engineer told colleagues, “we don’t do deals for data we already have or already possess,” the filing states. Employees also expected rights holders could remove books from Google Play or pursue legal action.
No court has ruled on the claims. The filing follows earlier copyright litigation over Gemini training and brings a separate challenge from major publishers.
Publisher files sit at the center of the dispute
Hachette Book Group, Cengage Learning, and Elsevier say they supplied books and articles through several Google services. Google Books and Google Play Books handled previews and ebook sales, while Google Scholar supported academic search.
The plaintiffs claimed that Google made additional copies during model training and carried copyrighted material into later Gemini versions. Separate allegations cover works collected from pirate libraries and paywalled sites, as well as removed or altered copyright-owner information.
In tests cited in the filing, Google’s AI tool produced a textbook table of contents, a reproduced passage, and a roughly 2,000-word summary of Turow’s Innocent.
Plaintiffs contend that responses of that depth could replace purchases or paid access. Requested remedies include damages, an injunction, disclosure of Gemini’s training sources, and court-supervised destruction of unauthorized copies.
Older contracts face an AI test
One question runs through the lawsuit: does permission to host, sell, index, or search content also cover AI training?
The publishers maintain that it does not. Their case asks the court to decide whether material supplied for one service can be reused to develop another without a separate license.
Universities, research groups, and companies that place protected material on outside platforms have reason to review their existing terms. Agreements written before generative AI became common may say little about model training or secondary use.
Google now has to defend how Gemini was trained, while courts decide how far permission to access protected work can extend.
Related News: Google Cloud’s survey suggests agentic AI is outgrowing the infrastructure many enterprises have in place today.