Publishers open new front in AI copyright wars

The suit, captioned Elsevier Inc. et al. v. Meta Platforms, Inc. and Mark Zuckerberg (Case No. 26-cv-3689), alleges willful infringement of millions of copyrighted works to train Meta's Llama large language models.

Until now, most lawsuits targeting generative AI developers had been brought by individual authors and creative groups, with the recent Bartz v. Anthropic class action being a notable example. The new filing represents a significant escalation: a coordinated front spanning trade, educational, and academic publishing, backed by the Association of American Publishers (AAP), which is supporting the plaintiffs on legal matters.

The 65-page complaint outlines six causes of action, including three counts of direct copyright infringement covering reproduction by torrenting, reproduction via web scrapes, and reproduction during model training, plus a separate count for distribution by torrenting. Zuckerberg is personally named in a contributory infringement claim, and Meta also faces a Digital Millennium Copyright Act (DMCA) claim for the alleged removal and alteration of copyright management information (CMI).

According to the complaint, Meta acquired training data through two parallel infringing pipelines. First, the company allegedly downloaded large volumes of web-scraped material—including the Common Crawl, CCNet, and C4 datasets—which themselves contained scraped content from notorious pirate sites such as Z-Library, OceanofPDF, and the subscription service Scribd. Second, and more aggressively, Meta is said to have torrented millions of books and journal articles directly from pirate repositories including LibGen, Anna's Archive, Sci-Hub, and the Books3 dataset compiled from the Bibliotik tracker.

The plaintiffs cite particularly striking figures: between April and July 2024 alone, Meta is alleged to have downloaded 134.6 terabytes of data and uploaded approximately 40.42 terabytes during torrenting—the latter equivalent to roughly twice the textual volume of the Library of Congress. In total, the complaint alleges Meta torrented more than 267 TB of pirated material. Internal communications cited in the filing describe Meta engineers masking corporate IP addresses to avoid being traced, and one employee reportedly characterized torrenting from a corporate laptop as something that "doesn't feel right."

A central narrative thread in the complaint concerns Meta's abandoned licensing efforts. According to the filing, between January and April 2023 Meta discussed scaling its dataset licensing budget from $17 million to $200 million and approached major publishers directly. The initiative was reportedly killed after being escalated to Zuckerberg himself. As Macmillan CEO Jon Yaged put it in the publishers' joint statement, "It is unconscionable that one of the world's most valuable companies chose to steal millions of works from creators for its own self-enrichment."

The complaint quotes a Meta employee summarizing the rationale for shelving the licensing track: licensing even a single book would, in the employee's view, undermine the company's ability to lean on a fair-use defense. The plaintiffs argue this exposes deliberate intent rather than inadvertent overreach.

Beyond the unauthorized ingestion of training data, the complaint makes the case that Llama's outputs themselves now harm the publishing market in five distinct ways: verbatim and near-verbatim reproductions, paraphrased substitutes, low-quality knockoffs, market-flooding through AI-generated books, and unauthorized derivative works. The filing describes the system as functioning, in effect, as an infinite substitution machine.

The complaint cites concrete examples of damage already occurring in the marketplace. One user reportedly generated a 100-chapter book from a single Llama 3.1 70B prompt. A self-published author who has released 171 books over seven years allegedly left a telltale AI prompt in a published volume. Another writer who issued three books in three months reportedly left in a prompt instructing the AI to rewrite passages to align more closely with a named established author's style. The complaint also points to a 2026 study finding that roughly three-quarters of titles in Amazon's "Success" self-help subgenre appear to have been written by AI.

Meta's financial exposure could be substantial. The complaint references Meta's own projections that AI products could generate between $460 billion and $1.4 trillion in revenue through 2035. Zuckerberg's personal net worth has reportedly climbed past $200 billion, growth the plaintiffs link directly to AI-driven gains following Llama's launch.

The plaintiffs are seeking statutory damages, actual damages and profits, attorneys' fees, and—crucially—injunctive relief that would order the destruction of all infringing copies in Meta's possession. The legal team includes Oppenheim + Zebrak LLP, Debevoise & Plimpton LLP, and Keller Rohrback L.L.P.

In a statement, Hachette CEO David Shelley said the publisher was proud to stand alongside Turow, a longstanding Hachette author and former Authors Guild president. Maria Pallante, president and CEO of the AAP, framed the action as a defense of intellectual property in the 250th year of the United States. The Authors Guild separately welcomed the suit, calling it another important step in holding AI developers accountable for what it characterizes as mass-scale theft of authors' works.

Beyond the named plaintiffs, the case is structured as a putative class action covering all owners of registered copyrights for ISBN-bearing books and DOI- or ISSN-bearing journal articles allegedly reproduced or distributed by Meta during the development of any Llama model. The class period begins in October 2022 and extends through the present.

The lawsuit lands at a moment when the publishing industry's posture toward generative AI is hardening, even as some houses simultaneously pursue licensing arrangements with cooperative AI companies. With Anthropic having recently settled its own author class-action and OpenAI defending against the New York Times and other plaintiffs, Meta now joins the front lines of what is rapidly becoming the defining legal battle over the foundations of generative AI.

For the publishing industry, the message embedded in the complaint is unambiguous: copyright law, the plaintiffs argue, is not a referendum on AI itself, but on what they describe as the technology's greedy and irresponsible deployment.