Publishers sue Google over Gemini AI training data

TL;DR:

• Hachette Book Group, Cengage Learning, Elsevier and author Scott Turow have sued Google in federal court in New York over books used to train Gemini.
• The suit alleges Google repurposed works supplied for limited services such as Google Books and Play Books without permission or payment.
• The complaint claims Google internally flagged exposure of “$10Bs-$100Bs in potential fines”.

Three major publishers and a bestselling author have accused Google of “one of the most prolific infringements of copyrighted materials in history”, filing suit in federal court in New York over the books used to build its Gemini models. The claimants — Hachette Book Group, Cengage Learning, Elsevier and author Scott Turow — argue Google took works supplied for narrow purposes and used them for something else entirely.

Supplied for snippets, used for training

The alleged mechanism matters more than the rhetoric. The publishers say Google repurposed books provided for limited services such as Google Books, Google Play Books and Google Scholar — services permitting specific uses like displaying searchable snippets or selling ebooks, but not, they argue, copying the works to train commercial AI. “Desperate to maintain its online dominance, Google abandoned its early motto of ‘Don’t be evil’,” the suit states. According to the complaint, Google made the copies despite internal discussions acknowledging the legal risk, flagging that it could face “$10Bs-$100Bs in potential fines” over texts provided for Play Books.

The competitive harm is put vividly. Gemini could generate “a 100-page murder mystery set in a quiet seaside town filled with secrets, that substitutes for an original copyrighted murder mystery on which Gemini trained” in 20 minutes for 39 cents. “No publisher or author can compete with that.” Named works include NK Jemisin’s The Fifth Season and Lemony Snicket’s Who Could That Be at This Hour?

The case joins a widening front. Authors and publishers have sued Google, OpenAI, Anthropic and Meta over training data, with mixed results: a judge ruled in Meta’s favour last June, while Anthropic settled for $1.5 billion with authors who alleged pirated copies of their books trained Claude. Thousands of authors including Kazuo Ishiguro, Philippa Gregory and Richard Osman published an “empty” book earlier this year in protest.

The contrast with the licensing route is stark, and it runs through a British company on the same day: Bloomsbury reported that AI licensing revenue is supporting its profit forecast under a partnership struck with Google last September. The same company, two models — one publisher paid, three suing.

Looking forward

The plaintiffs want statutory damages, a permanent injunction and a court order requiring Google to destroy unauthorised copies used in training. Google did not respond to a Guardian request for comment. The case follows an earlier attempt by Hachette and Cengage to join a 2023 suit against Google, which the company opposed — prompting this separate action.