Meta is staring down a fresh legal brawl in New York, as five heavyweight publishers and best-selling author Scott Turow accuse the tech giant of quietly copying millions of books and journal articles to fuel its Llama artificial intelligence models.
The lawsuit, filed Tuesday in Manhattan federal court, comes from Elsevier, Cengage, Hachette, Macmillan and McGraw Hill. They want a judge to let them represent a broad class of copyright owners and are asking for unspecified monetary relief, turning up the heat in a nationwide fight over whether AI developers can mine copyrighted text without permission.
According to Reuters, the proposed class action claims Meta "pirated" millions of works, from textbooks and scientific papers to children's stories and adult novels, to train Llama to answer user prompts. Maria Pallante, president of the Association of American Publishers, blasted the alleged "mass-scale infringement" as something other than public progress and warned that AI development should not lean on pirate sites over legitimate scholarship and imagination. The filing also asks the court to certify a class and to award damages and other relief.
The case lands in the middle of a wave of high-stakes copyright fights targeting AI companies and follows a headline-grabbing settlement last year: Anthropic agreed in 2025 to pay $1.5 billion to a group of authors, a deal that highlighted just how expensive these disputes can get for large language model developers, according to AP.
What the complaint alleges
Filed in the Southern District of New York, the complaint accuses Meta of copying millions of protected works and even calls out specific titles like N.K. Jemisin’s "The Fifth Season" and Peter Brown’s "The Wild Robot" as examples of material allegedly used without permission. The publishers are asking the court to grant class status and to award unspecified monetary damages along with injunctive relief, according to Reuters.
Legal landscape and earlier rulings
Federal judges have issued mixed rulings on whether training large language models on copyrighted books counts as fair use, leaving a patchwork of precedent that both sides are now testing in courts across the country. In one closely watched 2025 case, a San Francisco judge ruled for Meta and criticized the authors' legal strategy, a ruling in favor of Meta that publishers have been eyeing warily.
What’s at stake
If the publishers win class certification and ultimately prevail, AI companies could face statutory damages, injunctions and court orders to strip out unlicensed material or retrain models on properly licensed data. Those outcomes would carry serious commercial consequences. Researchers have also warned that large models can memorize and reproduce copyrighted passages, adding pressure on firms to audit and clean their training datasets, as explored in The Atlantic.
What to watch next
From here, expect early motions to dismiss, sharp fights over how much Meta must reveal about its training pipelines and a bruising debate over whether torrent-sourced or so-called "shadow library" content can be shielded by fair use. Because this is a brand-new filing in the Southern District of New York, months of briefing and evidence gathering are likely before any major rulings land.
The complaint dropped Tuesday in Manhattan federal court, and we will be watching the docket, the motions and the public spin from both sides as this one unfolds.
