Elsevier, Cengage, Hachette, Macmillan and McGraw Hill, along with author Scott Turow, filed a proposed class action in Manhattan on Tuesday, alleging that Meta pirated millions of his works to teach Llama. After Judge Chhabria’s ruling in June 2025, plaintiffs with stronger evidence of market harm waited their turn.
On Tuesday morning, five of the world’s largest publishers and one of America’s best-known authors entered a federal courthouse in Manhattan to file a proposed class action complaint against Meta Platforms.
This was reported by the Reuters agency Along with author Scott Turow, Elsevier, Cengage, Hachette, Macmillan and McGraw Hill allege that Meta has pirated millions of their books and journal articles to train large-tongued Llama models without permission, payment or license. The complaint asks the court to certify the case as a class action representing all similarly situated rights holders.
This is the latest in a long line of AI-training copyrights to date. In essence, it is meaningfully different from most of the previous ones.
Why this event is not Kadrey
Anyone who follows artificial intelligence copyright litigation will recognize this name in the nest of precedents: Kadrey v. Meta. This earlier case, filed in 2023 in the Northern District of California by authors including Sarah Silverman, Richard Kadrey, Christopher Golden, Ta-Nehisi Coates, Junot Díaz, and Michael Chabon, made essentially the same claims: Meta’s downloads of copyrighted books from pirated libraries (LibGen, Z-Library, and Armachis) used them.
Court records cited by Tom’s Hardware Meta employees were found to be torrenting approximately 82 terabytes of pirated material in the process. Mark Zuckerberg personally signed off on the use of LibGen for Llama training, despite internal AI executives labeling it a “dataset we know is pirated.” “Undermine (Meta’s) negotiating position with regulators.”
And Meta won the job. Judge Vince Chhabria in June 2025 Made the final decision for the meta on fair use grounds, found that using copyrighted books to teach Llama was transformative enough to overcome the fair use threshold. But Chhabria’s judgment was unusually narrow and unusually frank about its limits.
He made it clear that Meta’s victory “could be in significant tension with reality.” and that the decision applies only to the specific authors who filed the case. He made it clear that future plaintiffs could succeed if they presented stronger evidence of market harm, which the Kadrey plaintiffs wreathe in what he believed to be fair use analysis.
The document filed on Tuesday, on first inspection, reads as the exact type of work Chhabria invited.
What do publishers bring that authors haven’t?
There are three structural differences between Kadrey and the new claim, and all three favor the plaintiffs. The first is the directory. While Kadrey involved about 666 specific books by a small group of individual authors, the new complaint covers the entire publishing operations of five companies that together account for a significant portion of the world’s academic, educational and trade publications.
According to Reuters’ description of the complainttitles include not only literary works such as NK Jemisi’s The Fifth Season and Peter Brown’s Wild Robot, but also textbooks, scholarly journal articles, and reference works. The market for these works, particularly the academic and educational categories, is structurally different from the trade fiction market that dominates the Kadrey claimant set.
The second is evidence of market harm. Academic and educational publishers can document specific revenue streams replaced by AI-trained models in a way that individual authors typically cannot. When Llama answers a student’s biology question that requires consulting a Cengage textbook, the substitution is direct and measurable.
Plaintiffs, under the standard litigation strategy for this type of case, will present this replacement as quantifiable marketable damages, which Chhabria identified as missing from Kadrey in June. An analysis of recent fair use decisions by Reed Smith He noted that harming the market, rather than transformation, is now an operational legal battleground.
Third is the context of the licensing market. Since 2023, AI companies have signed an increasing number of license agreements with publishers. Meta himself Reuters, CNN, Fox News, People Inc. and signed contracts with USA Today. for content licensing.
The existence of these licenses is an important fact from the point of view of fair use law: courts investigating market tortfeasor already have evidence that a licensing market exists, that some publishers value and negotiate participation in it, and that Meta has chosen to participate in some markets and bypass others. New plaintiffs will argue that bypassing others while licensing them is evidence of bad faith.
An anthropic settlement in the background
Tuesday’s case goes against another piece of recent precedent. anthropic, in a settlement the Authors’ Guild described as clearly significantearlier this year, it agreed to pay the authors as part of a settlement of an Anthropic class action against Bartz over similar claims. The amount and terms of the settlement offer a hint of what artificial intelligence training can produce when copyright cases reach a financial settlement rather than an early final judgment.
TNW followed Anthropic’s broader commercial trajectory in parallel, through a $1.5 billion enterprise services joint venture, IPO preparations and model placement programs; The settlement of Bartz is, financially, a manageable dam in this fund. With a different fact pattern for meta and a prior summary-judgment win, the calculation is different.
However, the solution is only one possible outcome. Another, testing the Meta case as it tested Kadrey, is betting that the fair use defense will hold up against more substantial plaintiffs. The risk in this strategy is asymmetric. A second loss for publishers would virtually settle the question for the entire market: llama-style training on a pirate corporation is fair use even when the plaintiffs are a major industry. A second win would cost the company more in damages and structural protection than it would have avoided in the first case.
Meta’s broader legal spending trajectory
The new case sits within a broader legal landscape that Meta has been navigating for some time. TNW reported last week In a phase-two trial of Meta-New Mexico in Santa Fe, the state is seeking a $3.7 billion adolescent mental health fund linked to algorithm changes, age verification mandates and the company’s youth safety record.
TNW’s analysis earlier this year It noted that Meta’s growing child safety legal exposure could ultimately cost it more than its $145 billion artificial intelligence program. Meta’s Q1 2026 capital guidance now it’s $125 billion to $145 billion a year, an order of magnitude that makes the outcome of any litigation seem small in absolute terms, but it raises the question of how much concurrent legal exposure a company can accept without commercial consequences.
There is also a wider regulatory background. TNW covered Anthropic’s Mythos and Eurogroup’s parallel concerns about AI capability and access; it’s a different set of regulatory concerns than copyright, but it’s part of the same larger story of how AI companies’ commercial momentum collides with multiple categories of slower-moving legal infrastructure. The publishers’ appeal on Tuesday is a copyright example of that clash.
He asks what the job actually is
The narrow legal question is whether using pirated copyrighted material to teach Llama constitutes fair use under US copyright law. The broader question that the publishing industry is actually trying to address is whether existing fair use doctrine, written before generative models existed, can be expanded to accommodate them, or whether some new framework, statutory or judicial, needs to be established. The Kadrey ruling expanded the doctrine. The new case will test how far the stretch will go.
If the publishers win, even in part, the licensing market for AI training data becomes a structural fixture of the industry, and the material commercial implications for every model company now rely on a widely fractured corpus. If they lose, the practice of training pirated material at scale becomes legally sustainable in the United States, and the regulatory response shifts to legislatures rather than courts.
The procedure calendar will move slowly. Class certification, motions to dismiss, summary judgment briefing and trial planning will take 18 to 24 months in the normal course. Investing.com noted the broader market scanner context around the announcement of the suit, several other AI training copyright cases are now moving through US courts concurrently, and some may reach the appellate level before the matter is resolved. Tuesday’s appeal, in this sense, is not an immediate threat, but a long bet.
However, it is the most reliable long bet the publishing industry has placed against the AI training defendant. After Kadrey created what the Authors Guild called a “technical win” for Meta, but created a significant opening for future suitors, the industry has been waiting for the right slate of suitors to bring the next case. Tuesday records file names.
Over the next two years, Llama’s training corpus and, by extension, every comparable model built on similarly vast data will decide whether it is the original commercial sin of the AI era or its first widely accepted standard practice. There is no third conclusion that the courts can make.
Meta will argue, as it argued in Kadrey, that the use was transformative and no measurable market harm occurred. Publishers will argue that damages can be accurately measured through documentation, accounting, and licensing comparisons, and that Meta’s industry-wide selective licensing creates a market in which non-licensing of their works can be assessed. Justices Chhabria’s earlier decision actually wrote a brief for both sides.
The person who read the June opinion most closely, on the evidence available, sat down at the publishers’ desk on Tuesday.
