Meta Platforms Inc. CEO Mark Zuckerberg is facing explosive allegations that he personally directed his company to abandon licensing talks with publishers and instead “steal” millions of copyrighted books to train the company’s Llama artificial intelligence (AI) models.
In a class-action lawsuit filed Tuesday in Manhattan federal court, a heavyweight coalition of publishers — including Hachette, Macmillan, McGraw Hill, Elsevier, and Cengage — joined best-selling novelist Scott Turow in accusing the tech giant of systemic piracy. The complaint alleges that Meta bypassed established markets to populate its AI training sets with unlicensed works sourced from Anna’s Archive, a notorious search engine that indexes piracy repositories like LibGen and Sci-Hub.
The plaintiffs are seeking unspecified monetary damages and a permanent injunction to bar Meta from using their copyrighted works without authorization.
The legal filing illustrates a picture of executive-level misconduct, claiming Meta initially approached publishers to discuss licensing. However, the suit alleges that on Zuckerberg’s “personal instruction,” Meta walked away from the negotiating table to pursue a more “expedient” path: downloading the material illegally.
“That path deprived publishers and authors of fair compensation and spurned established licensing markets,” the lawsuit said.
The plaintiffs argue Meta’s Llama model doesn’t just learn from their data; it actively mimics it to create “inferior knockoffs” and “verbatim copies” that threaten the economic survival of the literary world. Examples cited in the filing include Llama generating a “passable knockoff” of Sylvia Day’s “One with You” and producing travel guides that convincingly mimic the distinct creative voice of author Becky Lomax.
The publishers also sounded the alarm over a “flourishing” market for AI-generated books on platforms like Amazon.com Inc., which they say are displacing human-authored works at an unprecedented volume. Maria Pallante, president of the Association of American Publishers, criticized the tech industry’s priorities, stating that companies are choosing “pirate sites over scholarship and imagination.”
Meta has signaled it will not back down. In a statement, a spokesperson defended the company’s practices, framing AI as a driver of global innovation. “Courts have rightly found that training AI on copyrighted material can qualify as fair use,” the spokesperson said. “We will fight this lawsuit aggressively.”
The case arrives amid a volatile legal landscape for generative AI. While some federal judges have sided with tech firms on fair use grounds, others have allowed piracy claims to proceed to trial. The stakes were recently heightened when Anthropic, a Meta competitor, reportedly reached a landmark $1.5 billion settlement to resolve a similar class-action suit involving the use of pirated digital books.
“Demand for training data is exploding across nearly every category, and the frontier labs are looking everywhere to find it. Candidly, that has been very good for our business,” Troveo CEO Marty Pesis said in an email. “You can often predict the quality of the model by understanding the quality and depth of data used to train it. Training on multidimensional workflow data from highly skilled technical workers is a recipe for building one of the world’s best models.”
