Google faces another AI training lawsuit from major publishers

Google built Gemini on a lie. That is the core accusation in a class-action complaint filed this week by Hachette, Cengage, Elsevier, author Scott Turow, and the writer collective S.C.R.I.B.E. The publishers didn't scrape the open web. They handed Google their catalogs under a specific contract: make our books searchable in Google Books. Snippets only. Bibliographic data. No full-text access. The lawsuit alleges Google took those same files, stripped the copyright metadata, and shoveled them into Gemini's training corpus anyway.

The detail that should keep Sundar Pichai awake at night isn't the scraping. It's the paper trail. Plaintiffs cite an internal Google document that allegedly describes using copyrighted books for AI training as "highly problematic for Google" and quantifies the exposure at "$10Bs-$100Bs in potential fines." That isn't negligence. That is willful infringement with a risk model attached. If the document holds up in discovery, Google's fair-use defense collapses into bad faith.

Fair use is the shield every AI company hides behind. Two California judges have already bought it, ruling that ingesting copyrighted works for model training fits a doctrine written before the internet existed. Those rulings are comfort food for Mountain View and Menlo Park. But they are not precedent. They are district-level opinions in a circuit that loves expansive fair-use readings. The Google case sits in the Southern District of New York. Different circuit. Different judges. Different factual substrate: a licensed corpus, not a scraped one. That distinction matters. Courts hate bait-and-switch.

Anthropic just wrote a $1.5 billion check — the largest copyright payout in U.S. history — to settle claims it pirated books for training. Roughly half a million writers qualified for a minimum of $3,000 each. Thousands opted out. They didn't opt out because the money was insulting. They opted out because a settlement buys silence. They want a ruling. They want a court to say: you cannot build a commercial model on our work without a license. That is the precedent the California decisions denied them.

Google's relationship with publishers makes this uglier than the OpenAI or Meta fights. Google Books was a partnership. Publishers trusted Google with their inventory because the product drove sales, not substitution. Gemini does the opposite. It synthesizes the book so the user never buys it. The lawsuit frames this as theft by conversion: Google took assets entrusted for discovery and repurposed them for displacement. That narrative resonates with juries. It also resonates with judges who see a fiduciary dimension in the licensing relationship.

The $10–100 billion figure in that internal memo is either hyperbolic or horrifying. At statutory damages of up to $150,000 per willful infringement, the math works fast. Millions of books times multiple copies times willfulness equals a number that exceeds Google's market cap. No court awards that. But the memo proves Google knew the legal geometry. It proceeded anyway. That converts a copyright case into a fraud-adjacent claim. Punitive damages enter the chat.

California's fair-use rulings rest on transformation: the model doesn't regurgitate the book; it learns patterns. But transformation is a spectrum. A search snippet transforms. A generative model that can reproduce chapters on command does not. The New York court will hear expert testimony on memorization rates, verbatim reproduction, and the commercial substitution effect. Google's own benchmarking likely shows Gemini can regurgitate. If plaintiffs subpoena those evals, the fair-use argument fractures.

This lawsuit won't settle quietly. The publishers who opted out of Anthropic's payout are the same cohort driving this case. They are organized, capitalized, and allergy-prone to NDAs. Google will fight because it must. A loss here doesn't just cost money; it forces a retraining regime that breaks the current economics of foundation models. The industry watches. The next licensing negotiation watches. The Supreme Court eventually watches.

Google's silence so far is standard counsel advice. But silence doesn't scrub the memo. It doesn't un-sign the Google Books agreements. And it doesn't move the venue back to the Ninth Circuit. The company bet that fair use would become settled law before any plaintiff reached a hostile jury. That bet looks shaky. The publishers just called it.