NEW YORK – Scott Turow spent four decades writing legal thrillers. This week, he became a plaintiff in one. The novelist joined a class action lawsuit filed Monday in U.S. District Court for the Southern District of New York by Hachette, Cengage, and Elsevier, along with the literary advocacy organization S.C.R.I.B.E., accusing Google of using copyrighted books to train its Gemini artificial intelligence platform without authorization or compensation.
The complaint rests on a specific factual claim. Publishers gave Google access to their books for Google Books, a limited-preview search tool designed to display only brief text snippets. The plaintiffs argue Google took that narrowly scoped access and fed the same materials into Gemini’s training pipelines – a use no licensing agreement covered. Books uploaded to Google Play Store under similarly restricted terms are named in the complaint as well. The full details of the filing were reported by TechCrunch.
“Google illegally copied works from all these scope-limited programs for AI training, knowing it lacked authorization to do so,” the complaint states. Beyond unauthorized use, the lawsuit accuses Google of intentionally removing or altering copyright information embedded in the training materials – an allegation that goes beyond negligent repurposing into deliberate concealment of the books’ origins.
What may prove most consequential is a document the plaintiffs cite from inside Google. According to the complaint, an internal Google document acknowledged that AI training on copyrighted books could be “highly problematic,” with potential financial exposure reaching “$10Bs-$100Bs in fines.” If authenticated through discovery, that document would establish that Google’s legal team assessed the liability long before any lawsuit arrived.
The lawsuit lands in a complicated legal moment. Earlier rulings from California federal courts found that training AI on copyrighted material constitutes fair use – a reading Google will almost certainly deploy in its defense. The Southern District of New York is not bound by those decisions. The plaintiffs’ argument is structurally different: they are not claiming all AI training on books is impermissible. They are claiming Google received access under specific, contractually limited terms and extended that access without consent – a contractual breach layered onto a copyright violation.
The financial benchmark that will shape any settlement discussion is Anthropic’s $1.5 billion payment to music publishers earlier this year – the largest copyright payout in U.S. history at the time. That settlement established real financial exposure for AI developers even when fair use arguments are technically available. Google’s situation differs in scale: a broader pool of affected works, deeper corporate resources, and now an internal document that may give plaintiffs’ counsel leverage in settlement talks. The earlier dispute over how internal communications shaped the Apple-OpenAI intellectual property dispute showed how quickly discovery documents can alter a company’s negotiating position.
Google has not responded publicly to the complaint. S.C.R.I.B.E. had not previously been a major litigant in AI copyright cases; the organization’s inclusion suggests the publisher coalition is assembling a broader front than the music industry’s campaign against Anthropic. How many additional publishing houses join the class as the case progresses will determine the scale of Google’s potential exposure – and whether the scope of training data at issue turns out to be larger than what the founding complaint describes.
For Turow and the authors the class now represents, the lawsuit is less about abstract legal principle than about whether publishers retain meaningful control over how their work enters AI training pipelines. The complaint does not argue that all AI training on published books is unlawful. It argues that Google was trusted with access under defined conditions it chose not to honor – and that the gap between what was permitted and what was done is the liability.
The question neither side can answer yet is whether the SDNY will treat scope-limited access agreements as legally distinguishable from open access, or whether California’s fair use ruling swallows those distinctions. The call by AI industry leaders for clearer governance frameworks has focused almost entirely on model behavior and frontier risk – not on what rights-holders are owed by systems their content shaped. This lawsuit may force the answer sooner than any voluntary standards body would have reached it.
