Google Faces Class Action Over Books Used To Train Gemini

A formidable coalition of publishers and authors has launched a proposed class-action lawsuit against Google, alleging the tech giant unlawfully copied millions of copyrighted books and journal articles to train its advanced artificial intelligence model, Gemini. The suit, filed in the U.S. District Court for the Southern District of New York, claims that Google utilized content sourced from its own platforms—Google Books, Play Books, and Scholar—as well as extensive web scrapes, including material from illicit pirate sites and paywalled subscription libraries, without obtaining the necessary permissions or providing compensation. This legal challenge ignites a pivotal debate over the boundaries of "fair use" in the age of generative AI and the fundamental rights of creators to control their intellectual property.

The Heart of the Allegations: Unauthorised Reproduction and DMCA Violations

The lawsuit, spearheaded by prominent entities including Hachette Book Group, Cengage Learning, Elsevier, renowned novelist Scott Turow, and his company S.C.R.I.B.E., was officially filed on July 10, with the Association of American Publishers (AAP) publicly announcing its support on the same day. The plaintiffs contend that the original agreements and purposes for which their works were provided to Google’s services did not extend to training commercial AI models. They assert that Google’s actions constitute a widespread and willful infringement of copyright.

Specifically, the complaint articulates four distinct counts against Google. Three of these charges center on unauthorized reproduction under the Copyright Act, addressing works obtained through:

1. Google’s proprietary services: This includes the vast digital libraries of Google Books, the e-commerce platform Play Books, and the academic search engine Google Scholar. Publishers argue that content supplied to these platforms was for specific distribution, viewing, or search purposes, not for foundational AI model training.
2. Extensive web scraping: The lawsuit alleges that Google’s data collection for Gemini extended to comprehensive web crawls, which reportedly included copyrighted materials found on pirate websites and behind the paywalls of subscription-based academic or news libraries. This method bypasses any direct agreements or permissions.
3. Copying during AI training: The very act of ingesting and processing these millions of works into Gemini’s neural networks is cited as an act of unauthorized copying.

The fourth count alleges that Google violated the Digital Millennium Copyright Act (DMCA) by removing or altering copyright management information (CMI) associated with the copied works. CMI typically includes copyright notices, author names, and other identifying information crucial for intellectual property protection. The plaintiffs are seeking substantial damages, a permanent injunction to prevent further unauthorized use, a detailed accounting of all copyrighted works used to train Gemini, and court orders mandating the deletion of any unauthorized copies.

Adding weight to their claims, the plaintiffs’ filing quotes what they describe as internal Google documents. One such document reportedly labeled the use of books from Google Play Books for AI training as "highly problematic for Google," cautioning about "potential fines from $10Bs-$100Bs." Another striking quote attributed to Gemini’s lead engineer allegedly states, "we don’t do deals for data we already have or already possess." While these documents are not public, and the quotes are presented by the plaintiffs, they suggest an internal awareness within Google regarding the contentious nature of their data sourcing practices. As of the time of publication, Google has not issued an official comment on the specific allegations in this complaint, and no court has yet ruled on any of the claims.

A Broader Battle: AI, Copyright, and the Future of Content

This lawsuit is not an isolated incident but rather a significant front in a rapidly expanding legal war between content creators and generative AI developers. The meteoric rise of large language models (LLMs) like Google’s Gemini, OpenAI’s ChatGPT, and Meta’s Llama has been fueled by their ability to process and learn from unprecedented volumes of human-created data, including books, articles, code, and images. The fundamental question at the heart of these disputes is whether this ingestion and transformation of copyrighted material for AI training constitutes "fair use" under existing copyright law or if it demands new forms of licensing and compensation for creators.

The concept of fair use, codified in Section 107 of the U.S. Copyright Act, allows for limited use of copyrighted material without permission for purposes such as criticism, comment, news reporting, teaching, scholarship, or research. Courts typically weigh four factors: the purpose and character of the use (especially whether it is transformative and non-commercial), the nature of the copyrighted work, the amount and substantiality of the portion used, and the effect of the use upon the potential market for or value of the copyrighted work. AI developers often argue that training an AI model is inherently transformative, as the model does not reproduce the original work directly but rather learns patterns and relationships from it to generate new content.

However, authors and publishers counter that this "learning" is entirely dependent on their original, expressive works, which represent years of creative effort and significant financial investment. They argue that if AI models can generate content that competes with or even replaces human-created works without proper licensing, it will devalue creative industries and undermine the livelihoods of creators. The AAP, in its announcement, underscored this sentiment, emphasizing the need for creators to be justly compensated and to maintain control over how their intellectual property is used in the digital age.

Google’s Stance and the Limits of Opt-Out Mechanisms

Google has previously articulated its position on AI training data. In a policy paper published on June 25, the company argued that training on publicly available web data constitutes a "transformative, non-expressive use" protected by fair use. Google also highlighted the existence of machine-readable controls, such as the Google-Extended robots.txt token, which websites can implement to opt out of having their content used for future Gemini training and some grounding uses. This mechanism allows webmasters to signal their preference regarding AI data collection.

However, the current lawsuit’s allegations complicate this narrative significantly. The plaintiffs contend that the materials in question — millions of books and articles — did not arrive through standard web crawling channels where robots.txt files would be relevant. Instead, they were reportedly sourced via two primary routes:

1. Direct Agreements for Google Services: The books were supplied directly to Google through various agreements for services like Google Books, Play Books, and Scholar. In these scenarios, a robots.txt file is irrelevant because the content was provided directly under specific contractual terms. The core of the dispute here is whether these initial agreements implicitly or explicitly granted permission for AI training, which the publishers vehemently deny.
2. Web Scrapes from Non-Standard Sources: The complaint also points to copies allegedly appearing in Common Crawl after being hosted on pirate sites and subscription libraries. Since these unauthorized or paywalled copies are hosted on domains not directly controlled by the original copyright holders, their robots.txt files cannot regulate Google’s alleged scraping of these illicit or restricted sources. This highlights a significant loophole where content acquired outside legitimate channels could be ingested for AI training, further aggravating copyright holders.

The Google-Extended token, therefore, becomes a smaller factor in this particular dispute. While BuzzStream data from January indicated that 79% of top news sites block at least one AI training bot, this mechanism primarily addresses content obtained through standard web crawling. The lawsuit against Google focuses on content allegedly acquired through channels that these conventional crawler settings simply do not affect. This reinforces the broader argument made by groups like Digital Content Next (DCN), which, last month, sent a cease and desist letter to the Common Crawl Foundation, asserting that copyright law does not operate as an opt-out system; rather, it requires explicit permission for use.

Chronology of Key Developments:

• Early 2000s: Google launches the Google Books project, scanning millions of books, leading to a decade-long legal battle over copyright, eventually largely resolved in Google’s favor under fair use for snippet display.
• Late 2010s – Early 2020s: Rapid development and deployment of generative AI models, including Google’s Gemini, relying heavily on vast datasets.
• 2025 (California Rulings): Two significant Northern California District Court rulings address AI training and fair use. In the Anthropic case, the court denied summary judgment on pirated central-library copies, indicating potential liability. In the Meta case, the judge ruled in favor of Meta regarding fair use for training on certain records, but stressed the decision’s specificity to those particular plaintiffs and their record, avoiding a broad precedent.
• June 25 (Current Year): Google publishes a policy paper defending AI training on public web data as "transformative, non-expressive use" under fair use protections and highlighting opt-out mechanisms.
• Last Month (Prior to Lawsuit): Digital Content Next sends a cease and desist letter to the Common Crawl Foundation, asserting that copyright law mandates an opt-in system, not an opt-out.
• July 10 (Current Year): Hachette Book Group, Cengage Learning, Elsevier, Scott Turow, and S.C.R.I.B.E. file the proposed class-action lawsuit against Google in the U.S. District Court for the Southern District of New York. The Association of American Publishers announces its support.

Implications for the AI Industry, Copyright Law, and Creators

The outcome of this lawsuit could have profound implications for the entire generative AI industry, the interpretation of copyright law, and the future economic models for creators and publishers.

• For the AI Industry: A ruling against Google could necessitate a fundamental shift in how AI models are trained. It might compel AI developers to seek explicit licenses for all copyrighted material, leading to potentially massive licensing costs, a scramble for alternative, royalty-free datasets, or even a slowdown in AI development. Conversely, a ruling in Google’s favor could solidify the "transformative use" argument for AI training, potentially reducing the legal burden on developers but further alienating content creators.
• For Copyright Law: This case, alongside others currently unfolding, will test the adaptability of existing copyright frameworks to new technological paradigms. Courts will grapple with defining what constitutes "copying" when an AI model processes and learns from data, and how the "transformative" nature of AI output should be weighed against the "market harm" to original works. It could lead to legislative calls for new copyright laws specifically tailored to AI, similar to how the DMCA addressed digital content.
• For Authors and Publishers: A favorable ruling for the plaintiffs could establish a precedent for fair compensation and control over their works in the AI era. It would empower creators to demand licensing fees, potentially creating new revenue streams for industries struggling with digital disruption. Conversely, a loss could further diminish their control and financial leverage, raising concerns about the long-term viability of creative professions if their foundational works can be freely exploited to train competing AI systems.

The plaintiffs, notably, chose to file in New York after initially considering intervention in the ongoing "In re Google Generative AI Copyright Litigation" in California. This decision suggests a strategic move to preserve specific claims they believe fall outside the scope of that proposed class action, indicating a multifaceted legal approach to the burgeoning challenges of AI and copyright.

The next immediate step in this high-stakes legal drama will be Google’s formal response to the complaint, which could take the form of an answer addressing the allegations directly or a motion to dismiss, arguing legal insufficiencies in the plaintiffs’ case. Regardless of the immediate procedural outcome, this lawsuit marks a critical juncture in defining the relationship between groundbreaking AI technology and the foundational principles of intellectual property. The legal system is once again tasked with balancing innovation with the protection of creative rights, in a conflict that will undoubtedly shape the digital landscape for decades to come.