Publishers Sue Meta Over Pirated Books Used to Train Llama

A coalition of major publishers has filed a sweeping class action lawsuit against Meta Platforms, alleging the tech giant systematically harvested millions of copyrighted books and journal articles from pirate websites to train its Llama large language models. The complaint names CEO Mark Zuckerberg as a defendant, accusing him of personally approving and enabling what plaintiffs call 'massive copyright infringement.'

The lawsuit, filed by publishers Cengage Learning, Hachette Book Group, Macmillan Publishers, McGraw-Hill, and bestselling author Scott Turow, demands a jury trial and seeks damages for what it describes as one of the largest acts of literary theft in modern history.

Key Takeaways

• Who is suing: Cengage Learning, Hachette, Macmillan, McGraw-Hill, and author Scott Turow have filed a class action suit against Meta
• The allegation: Meta allegedly scraped millions of copyrighted books and academic articles from piracy websites to train its Llama AI models
• Zuckerberg named personally: The complaint accuses the CEO of directly approving and encouraging the infringing activity
• Copyright stripping: Meta allegedly removed copyright management information from the works before feeding them into training pipelines
• Meta's defense: The company plans to argue that using copyrighted content for AI training constitutes fair use under U.S. law
• Jury trial requested: Publishers are pushing for a full jury trial, signaling confidence in the strength of their evidence

Publishers Accuse Meta of Systematic Piracy

The lawsuit paints a damning picture of Meta's data acquisition practices. According to the complaint, Meta did not simply stumble upon copyrighted content — it actively sought out pirated copies of books and academic publications from well-known illegal download sites. The publishers allege that Meta's engineers knowingly accessed these repositories to build massive training datasets for the Llama family of models.

What makes this case particularly notable is the allegation that Meta deliberately stripped copyright management information (CMI) from the works before using them. Under the Digital Millennium Copyright Act (DMCA), removing or altering CMI is itself a violation of federal law, separate from the underlying copyright infringement. This means Meta could face liability on multiple legal fronts even if it manages to mount a partial fair use defense.

The publishers estimate that millions of individual works were affected, spanning fiction, nonfiction, textbooks, academic journals, and professional reference materials. The scale of the alleged infringement is staggering and, if proven, would represent one of the most extensive cases of unauthorized use of copyrighted material in the history of the technology industry.

Zuckerberg Personally Named as Defendant

Perhaps the most aggressive aspect of the lawsuit is the decision to name Mark Zuckerberg individually as a defendant. The complaint alleges that Zuckerberg did not merely oversee a company that happened to infringe copyrights — he personally approved and encouraged the practices in question.

Naming a CEO personally in a copyright suit of this magnitude is relatively unusual and suggests that the plaintiffs believe they have evidence of direct involvement. In most corporate litigation, executives are shielded by the corporate structure. By piercing that veil, the publishers are signaling that they intend to hold Zuckerberg accountable at the highest level.

This strategy also raises the stakes considerably for Meta's legal team. A finding of willful infringement — particularly with executive-level knowledge — could dramatically increase statutory damages. Under U.S. copyright law, willful infringement can carry penalties of up to $150,000 per work infringed. With millions of works allegedly involved, the potential liability is astronomical.

The Fair Use Battleground

Meta's spokesperson has already previewed the company's defense, stating that Meta intends to 'vigorously defend' itself and arguing that using copyrighted content to train AI models constitutes fair use. This defense has become the central legal question of the generative AI era, and no definitive court ruling has yet settled the matter.

The fair use doctrine, codified in Section 107 of the U.S. Copyright Act, allows limited use of copyrighted material without permission under certain conditions. Courts evaluate 4 factors:

• Purpose and character of the use: Is it transformative? Commercial use weighs against fair use
• Nature of the copyrighted work: Creative works receive stronger protection than factual ones
• Amount used: Using entire works (as alleged here) generally weighs against fair use
• Market impact: Does the use harm the market for the original work or potential licensing revenue?

Meta will likely argue that AI training is inherently transformative — the model does not reproduce the original text but rather learns patterns from it. However, the publishers counter that Llama-powered products directly compete with their works and that Meta's refusal to license content undermines the entire market for training data.

This case joins a growing roster of AI copyright disputes, including the New York Times v. OpenAI lawsuit filed in late 2023 and multiple suits brought by visual artists against Stability AI and Midjourney. Each case tests slightly different aspects of fair use, but together they are building toward what many legal experts expect will be a landmark Supreme Court ruling within the next few years.

How This Fits Into the Broader AI Copyright Crisis

The publishing industry's lawsuit against Meta is not happening in isolation. It reflects a rapidly escalating conflict between content creators and AI companies that has been building since the release of ChatGPT in November 2022.

Several parallel developments are shaping this landscape:

• The New York Times sued OpenAI and Microsoft in December 2023, alleging their models can reproduce Times articles nearly verbatim
• Getty Images filed suit against Stability AI for using its photo library without permission to train image generation models
• The Authors Guild organized thousands of writers, including John Grisham, Jodi Picoult, and George R.R. Martin, in separate litigation against OpenAI
• Music publishers have begun exploring similar legal action against companies training models on song lyrics and compositions
• The U.S. Copyright Office launched a formal inquiry into AI and copyright in 2023, but has yet to issue definitive guidance

Meta occupies a unique position in this debate because its Llama models are open-source (or more precisely, openly licensed). Unlike OpenAI's GPT-4 or Google's Gemini, which are accessible only through APIs, Llama weights are freely downloadable. This means that any copyrighted material baked into the model's training data is effectively distributed to every developer who downloads and deploys Llama — a fact the publishers are likely to emphasize in court.

What This Means for the AI Industry

The outcome of this lawsuit could reshape how every AI company approaches training data. If Meta loses, the implications extend far beyond a single company.

For AI developers, a ruling against fair use would mean that training on copyrighted content requires explicit licensing agreements. This would dramatically increase the cost of building large language models and could create significant barriers to entry for smaller players. Companies like OpenAI, Google, and Anthropic — all of which have trained on copyrighted material — would need to renegotiate their data strategies.

For publishers and creators, a favorable ruling would establish a clear legal right to compensation when their works are used for AI training. This could create an entirely new revenue stream — potentially worth billions of dollars annually — through training data licensing agreements. Some publishers, like those working with Axel Springer and Associated Press, have already struck deals with OpenAI, but many others have held out.

For businesses using Llama, the lawsuit introduces uncertainty. Companies that have built products on Meta's open-source models may face questions about the legal provenance of the underlying training data. While downstream users are unlikely to face direct liability, the reputational and regulatory risks are real.

Looking Ahead: A Defining Moment for AI and Copyright

This case is poised to become one of the most closely watched legal battles in the technology sector. Several factors will determine its trajectory in the months ahead.

First, the discovery phase could prove explosive. If the publishers obtain internal Meta communications — including any emails or messages from Zuckerberg — showing awareness that training data was sourced from pirate sites, the fair use defense becomes significantly harder to sustain. The allegation that CMI was deliberately stripped suggests the plaintiffs already have some evidence of intentional conduct.

Second, the case's timing matters. With the New York Times v. OpenAI proceeding in parallel, and multiple other AI copyright cases winding through federal courts, there is a growing possibility that circuit splits will emerge, ultimately forcing the U.S. Supreme Court to weigh in. A Supreme Court ruling on AI training and fair use would set precedent for the entire industry.

Third, Congress may intervene before the courts reach a final verdict. Several legislative proposals addressing AI and copyright are circulating on Capitol Hill, including bills that would require AI companies to disclose their training data sources and obtain consent from rights holders.

For now, the battle lines are drawn. The publishing industry — one of the oldest content businesses in the world — is challenging one of the newest and most powerful technology companies over a fundamental question: who owns the knowledge that powers artificial intelligence? The answer will define the economics of AI for a generation.