Hachette, Elsevier and Cengage sue Google, alleging Gemini AI trained on millions of pirated books

The plaintiffs and the filing

Three of the world's largest publishers (Hachette Book Group, Cengage Learning, and Elsevier) together with crime writer Scott Turow and his company S.C.R.I.B.E. filed a class action lawsuit against Google on Tuesday in the U.S. District Court for the Southern District of New York. The complaint accuses the tech giant of reproducing millions of copyrighted books, textbooks, and journal articles without permission to train its Gemini AI models. The plaintiffs argue that Google obtained these works through scope-limited programmes such as Google Books, Google Play Books, and Google Scholar, which allowed only specific uses like displaying searchable snippets or selling ebooks, not bulk copying for commercial AI training.

Google reproduced millions of copyrighted works without permission, without providing any compensation to authors or publishers, and with full knowledge that its conduct violated copyright law.

What Gemini can allegedly produce

The lawsuit claims Gemini can generate a 100-page murder mystery in 20 minutes for $0.39, a product the publishers say no human author or traditional publisher can compete with. The filing states that Gemini produces verbatim and near-verbatim copies of portions or entire works, replacement chapters of academic textbooks, summaries and alternative versions of famous novels, and inferior knockoffs that copy creative elements of originals. It also alleges that Gemini tailors outputs to mimic the expressive elements and creative choices of specific authors. Specific titles named in the complaint include N.K. Jemisin's The Fifth Season and Lemony Snicket's Who Could That Be at This Hour?.

The scale and speed at which Gemini can create books and compete with human writers is unprecedented, and it can only do that because Google copied Plaintiffs' and the Class's works to train its AI.

Google's internal knowledge and alleged concealment

The publishers claim Google's infringement was willful. The complaint cites an internal Google document acknowledging the company could face "$10Bs-$100Bs in potential fines" for using texts provided by publishers for Google Play Books. The plaintiffs also allege that Google stripped copyright management information from the works to conceal its training sources and facilitate unauthorised use. The suit characterises the episode as "one of the most prolific infringements of copyrighted materials in history," stating that Google abandoned its early motto of "Don't be evil" in a bid to maintain online dominance.

The broader legal landscape

This is the latest in a wave of copyright lawsuits against AI developers. In May, the same group of plaintiffs (Hachette, Cengage, Elsevier, and Turow) sued Meta on similar grounds in a New York court. A separate group of authors reached a $1.5 billion settlement with Anthropic in 2025 over pirated training materials for its Claude chatbot, though a judge later rejected the settlement as incomplete. Around half a million writers were eligible for payments of at least $3,000 under that deal, but many opted out to pursue further litigation. Meta won a partial victory last year when a San Francisco judge ruled its use of copyrighted materials constituted fair use, in a case brought by comedian Sarah Silverman and author Ta-Nehisi Coates. Two early California court decisions have favoured AI companies on fair use grounds, but the New York filing gives a different judge the opportunity to weigh in on the nuanced conflict.

What the plaintiffs want

Hachette is the third-largest book publisher in the United States behind Penguin Random House and HarperCollins (the latter signed a licensing deal with Microsoft in 2024 to supply books for AI training). Cengage is a major education publisher, and Elsevier is an academic publisher of journals including The Lancet and Cell. The plaintiffs are requesting an injunction and unspecified damages. They argue that Google could have licensed the content properly but chose not to, and that absent effective guardrails, Gemini will continue to produce outputs that substitute for the copyrighted works on which it was trained.