Hachette Book Group, Cengage Learning, Elsevier, novelist Scott Turow and his company S.C.R.I.B.E. have sued Google, alleging that millions of copyrighted books and journal articles were copied without permission to train its Gemini AI models. The proposed class action seeks statutory damages, an injunction against the alleged infringement and the destruction of unauthorized training copies.
The case is Hachette Book Group, Inc. et al. v. Google LLC, case number 1:26-cv-05870 in the U.S. District Court for the Southern District of New York. Although the publishers publicly announced it on July 13 and news reports followed on July 14, the federal docket records the complaint as filed on July 10, 2026.
As detailed by Hachette and reported by The Guardian and Publishers Weekly, the plaintiffs say Google repurposed works supplied for Google Books, Google Play Books and Google Scholar. Those arrangements allegedly permitted limited activities such as displaying searchable snippets, distributing ebooks and supporting scholarly discovery—not using complete works as raw material for commercial generative AI.
Google had not publicly responded to the allegations at the time of reporting.
The central accusation is more specific than the familiar claim that an AI company indiscriminately scraped copyrighted material from the web. The publishers allege Google already held high-quality digital copies because rightsholders had provided them for clearly defined services.
That distinction could become pivotal. The complaint argues that permission to sell an ebook or show a search snippet does not automatically authorize Google to reproduce the same work in a separate training pipeline for Gemini.
The plaintiffs also accuse Google of obtaining material through unauthorized web scrapes, including content from alleged pirate sources and behind paywalls. They claim works were copied repeatedly during model development and that copyright management information was stripped from some material to obscure its origin.
Among the books identified in reporting on the complaint are N.K. Jemisin’s The Fifth Season and Lemony Snicket’s Who Could That Be at This Hour? The proposed class, however, could extend far beyond trade fiction, covering educational books, scholarly articles, memoirs, poetry and other published writing.
Google successfully defended the scanning and indexing behind Google Books in earlier litigation. In that dispute, courts found the service’s searchable index and limited snippets to be transformative fair use rather than a substitute for the books themselves.
The publishers are trying to draw a firm boundary between that search function and Gemini. Google Books directs a user toward an existing work; the complaint argues that a generative model can produce new text that competes with it.
According to the plaintiffs, Gemini can generate summaries, textbook material, alternate versions of stories and prose reflecting the expressive choices of named authors. They argue that these outputs can displace legitimate sales while weakening the emerging market in which publishers license archives to AI developers.
One example described in the filing estimates that Gemini could create a 100-page murder mystery in roughly 20 minutes at a computing cost of 39 cents. That figure is an allegation intended to illustrate the publishers’ economic case, not an independently established retail comparison, but its purpose is clear: the plaintiffs want the court to see Gemini as a production system capable of creating low-cost substitutes.
That may prove more consequential than whether an ordinary user can coax Gemini into reproducing a particular passage word for word. Copyright law’s fair-use analysis considers effects on the potential market, and the publishers argue that synthetic books and educational materials threaten both sales of existing works and future AI licensing revenue.
Google is likely to dispute the characterization of generated material as a substitute and may argue that model training is transformative rather than expressive republication. Those defenses have gained some support in other AI cases, but outcomes have depended heavily on how developers acquired and stored their source material.
The lawsuit therefore combines several theories that should not be treated as interchangeable. Training on legitimately obtained books, retaining unauthorized copies, using allegedly pirated datasets, removing rights information and generating infringing output can raise different legal questions even when they occur in the same technical pipeline.
The plaintiffs say Google also recognized that publishers would probably view large-language-model training on their books as copyright infringement, withdraw content from Google Play Books or sue. Other alleged internal concerns included restrictive partner licenses, publisher sensitivity and heightened risk around a fair-use defense.
These statements remain allegations drawn from the plaintiffs’ filing. Their authenticity, context and legal significance will have to be tested through the court process, and a risk estimate inside a company is not itself proof that infringement occurred.
They could still matter to the question of willfulness. Statutory damages can rise when infringement is found to have been deliberate, while evidence that decision-makers understood and accepted a known legal risk can influence both damages and settlement pressure.
The plaintiffs are requesting more than compensation. They want a permanent injunction and an order requiring Google to destroy unauthorized copies used in training, remedies that could pose difficult technical questions about datasets, checkpoints and models derived from disputed material.
A court would have to decide what destruction means once information has passed through multiple preparation and training stages. Removing source files is straightforward compared with determining whether a trained model embodies protected expression and, if so, whether retraining is necessary.
For administrators and procurement teams, the lawsuit reinforces the need to separate training-data risk from the security and privacy controls normally reviewed during an AI deployment. A service may provide strong tenant isolation, identity management and data-loss-prevention features while still facing unresolved claims about its foundational datasets.
Organizations using generative AI to create customer-facing publications, course material, documentation or long-form commercial content should preserve records of prompts, outputs, model versions and human review. They should also examine vendor terms covering intellectual-property indemnification, output ownership and responses to infringement claims.
No Windows configuration change is required because of the filing, and there is no indication that Gemini access is about to disappear. The nearer-term consequence will be contractual: publishers may become less willing to supply digital archives without explicit AI restrictions, while model vendors face greater pressure to disclose licensing arrangements and provide stronger legal protections to customers.
The New York court must first address class certification, Google’s response and the publishers’ underlying infringement theories. The decisive issue will be whether access granted for Google Books, Google Play Books and Google Scholar also gave Google room to build Gemini—or whether the company turned a limited publishing relationship into an unauthorized AI training library.
The case is Hachette Book Group, Inc. et al. v. Google LLC, case number 1:26-cv-05870 in the U.S. District Court for the Southern District of New York. Although the publishers publicly announced it on July 13 and news reports followed on July 14, the federal docket records the complaint as filed on July 10, 2026.
As detailed by Hachette and reported by The Guardian and Publishers Weekly, the plaintiffs say Google repurposed works supplied for Google Books, Google Play Books and Google Scholar. Those arrangements allegedly permitted limited activities such as displaying searchable snippets, distributing ebooks and supporting scholarly discovery—not using complete works as raw material for commercial generative AI.
Google had not publicly responded to the allegations at the time of reporting.
Google Books Becomes the Legal Fault Line
The central accusation is more specific than the familiar claim that an AI company indiscriminately scraped copyrighted material from the web. The publishers allege Google already held high-quality digital copies because rightsholders had provided them for clearly defined services.That distinction could become pivotal. The complaint argues that permission to sell an ebook or show a search snippet does not automatically authorize Google to reproduce the same work in a separate training pipeline for Gemini.
The plaintiffs also accuse Google of obtaining material through unauthorized web scrapes, including content from alleged pirate sources and behind paywalls. They claim works were copied repeatedly during model development and that copyright management information was stripped from some material to obscure its origin.
Among the books identified in reporting on the complaint are N.K. Jemisin’s The Fifth Season and Lemony Snicket’s Who Could That Be at This Hour? The proposed class, however, could extend far beyond trade fiction, covering educational books, scholarly articles, memoirs, poetry and other published writing.
Google successfully defended the scanning and indexing behind Google Books in earlier litigation. In that dispute, courts found the service’s searchable index and limited snippets to be transformative fair use rather than a substitute for the books themselves.
The publishers are trying to draw a firm boundary between that search function and Gemini. Google Books directs a user toward an existing work; the complaint argues that a generative model can produce new text that competes with it.
The Complaint Targets Outputs as Well as Inputs
AI copyright disputes frequently focus on the moment data enters a training system: where a model developer obtained a file, whether it made a copy and whether that copying qualifies as fair use. The new Google case adds a market argument built around what Gemini can produce after training.According to the plaintiffs, Gemini can generate summaries, textbook material, alternate versions of stories and prose reflecting the expressive choices of named authors. They argue that these outputs can displace legitimate sales while weakening the emerging market in which publishers license archives to AI developers.
One example described in the filing estimates that Gemini could create a 100-page murder mystery in roughly 20 minutes at a computing cost of 39 cents. That figure is an allegation intended to illustrate the publishers’ economic case, not an independently established retail comparison, but its purpose is clear: the plaintiffs want the court to see Gemini as a production system capable of creating low-cost substitutes.
That may prove more consequential than whether an ordinary user can coax Gemini into reproducing a particular passage word for word. Copyright law’s fair-use analysis considers effects on the potential market, and the publishers argue that synthetic books and educational materials threaten both sales of existing works and future AI licensing revenue.
Google is likely to dispute the characterization of generated material as a substitute and may argue that model training is transformative rather than expressive republication. Those defenses have gained some support in other AI cases, but outcomes have depended heavily on how developers acquired and stored their source material.
The lawsuit therefore combines several theories that should not be treated as interchangeable. Training on legitimately obtained books, retaining unauthorized copies, using allegedly pirated datasets, removing rights information and generating infringing output can raise different legal questions even when they occur in the same technical pipeline.
An Alleged Internal Warning Raises the Stakes
The most damaging claim may concern what Google reportedly knew before or during Gemini’s development. The complaint cites internal analysis describing the use of publisher-provided copyrighted books as “highly problematic” and warning of “$10Bs-$100Bs in potential fines.”The plaintiffs say Google also recognized that publishers would probably view large-language-model training on their books as copyright infringement, withdraw content from Google Play Books or sue. Other alleged internal concerns included restrictive partner licenses, publisher sensitivity and heightened risk around a fair-use defense.
These statements remain allegations drawn from the plaintiffs’ filing. Their authenticity, context and legal significance will have to be tested through the court process, and a risk estimate inside a company is not itself proof that infringement occurred.
They could still matter to the question of willfulness. Statutory damages can rise when infringement is found to have been deliberate, while evidence that decision-makers understood and accepted a known legal risk can influence both damages and settlement pressure.
The plaintiffs are requesting more than compensation. They want a permanent injunction and an order requiring Google to destroy unauthorized copies used in training, remedies that could pose difficult technical questions about datasets, checkpoints and models derived from disputed material.
A court would have to decide what destruction means once information has passed through multiple preparation and training stages. Removing source files is straightforward compared with determining whether a trained model embodies protected expression and, if so, whether retraining is necessary.
Enterprise AI Buyers Inherit the Uncertainty
The immediate case concerns Google and publishers, but the practical issue reaches organizations deploying Gemini, Microsoft Copilot, OpenAI models, Anthropic Claude and Meta Llama from Windows PCs and cloud applications. IT departments are buying access to AI outputs without receiving a complete inventory of every work used during training.For administrators and procurement teams, the lawsuit reinforces the need to separate training-data risk from the security and privacy controls normally reviewed during an AI deployment. A service may provide strong tenant isolation, identity management and data-loss-prevention features while still facing unresolved claims about its foundational datasets.
Organizations using generative AI to create customer-facing publications, course material, documentation or long-form commercial content should preserve records of prompts, outputs, model versions and human review. They should also examine vendor terms covering intellectual-property indemnification, output ownership and responses to infringement claims.
No Windows configuration change is required because of the filing, and there is no indication that Gemini access is about to disappear. The nearer-term consequence will be contractual: publishers may become less willing to supply digital archives without explicit AI restrictions, while model vendors face greater pressure to disclose licensing arrangements and provide stronger legal protections to customers.
The New York court must first address class certification, Google’s response and the publishers’ underlying infringement theories. The decisive issue will be whether access granted for Google Books, Google Play Books and Google Scholar also gave Google room to build Gemini—or whether the company turned a limited publishing relationship into an unauthorized AI training library.
References
- Primary source: Social Samosa
Published: 2026-07-15T06:59:37.399000+00:00
Publishers allege Google illegally used copyrighted books to build Gemini AI
According to the complaint, books licensed to Google Books, Play Books, and Scholar were intended for snippets and ebook sales, but Google allegedly copied them to train its commercial AI models.
www.socialsamosa.com
- Independent coverage: informat.ro
Published: 2026-07-15T06:39:09+00:00
The major publishers are suing Google for the illegal use of protected books to train the Gemini artificial intelligence.
Three major publishers, Hachette Book Group, Cengage Learning, and Elsevier, along with writer Scott Turow, have sued Google in a federal court in New York, accusing the company of copyright infringement through the illegal use of millions of protected books to train the artificial intelligence...
informat.ro
- Independent coverage: Exchange4Media
Published: 2026-07-15T05:48:48+00:00
Publishers sue Google, allege Gemini was trained on copyrighted books without permission
A coalition of publishers and author Scott Turow sues Google, claiming unauthorized use of copyrighted books to train its Gemini AI, seeking damages and an injunction.
www.exchange4media.com
- Independent coverage: aol.co.uk
Published: 2026-07-15T00:51:42+00:00
Publishers sue Google over alleged use of books to train AI models - AOL
Book publishers sued Google on Tuesday, accusing the tech giant of illegally using copyrighted works to train its artificial intelligence models and generate content that competes with human authors. The lawsuit marks the latest legal battle over how AI developers use books and other creative...www.aol.co.uk - Independent coverage: The Cryptonomist
Published: 2026-07-14T23:03:02+00:00
Google Gemini Lawsuit Challenges AI Training Copyright Use
The Google Gemini lawsuit accuses the company of unauthorized use of copyrighted books for AI training, raising key fair use legal questions.
en.cryptonomist.ch
- Independent coverage: The Guardian
Published: 2026-07-14T18:16:56+00:00
Book publishers sue Google for copyright infringement over Gemini AI training | Books | The Guardian
Group of major publishers accuses the tech giant of ‘one of the most prolific infringements of copyrighted materials in history’www.theguardian.com
