Authors Sue Meta Over the Books It Bought

SAN FRANCISCO, Oct. 19 — AuthorsAuthors have sued Meta before over books it downloaded from “shadow libraries” online. On Monday they sued it over books it bought.

Five near-identical complaints filed in federal court in San Francisco on behalf of 292 plaintiffs repeat charges made in earlier suits and add a new one: that Project Vellum, under which, the complaints say, Meta bought print books en masse and scanned them to train its Muse models, was itself an infringement of copyright. Two firms, Stris & Maher and Freedman Normand Friedland, brought the suits; Bloomberg Law, citing their lawyers, put the number of authors at around 300.

The claim reaches past Meta: it asks whether buying a book and scanning it is a lawful way to gather text for training, a question on which the only ruling so far, by one tracker’s count, went the developer’s way. And the bill would be counted book by book. The authors seek statutory damages, which can reach $150,000 for each work infringed willfully, and they have chosen to press individual claims rather than join a class.Statutory DamagesThe Copyright Act lets a court award $750 to $30,000 for each work infringed, as it considers just, and up to $150,000 if the infringement was willful. It may cut the award to $200 where an infringer had no reason to believe it was infringing.

“Plaintiffs desire to retain full control of their case and avoid having their rights diluted by being swept into sprawling class-action settlements structured to resolve claims for pennies on the dollar,” the complaints say. Anthropic’s settlement with a class of authors, approved in July, set up a $1.5 billion fund, which works out to about $3,000 a work before costs and fees, according to the settlement’s website.Bartz v. AnthropicJudge William Alsup denied Anthropic summary judgment on the pirated copies in its central library and ordered a trial on them. The case settled instead; the court gave final approval on July 20, and the first payments are expected by Nov. 15.

The Scans

The complaints build the Vellum claim largely from Meta’s own answers to written questions, served on Sept. 4 in another suit, brought by Chicken Soup for the Soul. They say Meta began Vellum in or around mid-2025, “after it had been sued repeatedly for training Llama on pirated books,” and call it “an industrial-scale operation.” By Meta’s account, as the complaints quote it, the books were “first scanned into digital form and then run through OCR,” or optical character recognition, and the text was “then de-duplicated, processed and cleaned.”

Meta “has admitted that books it acquired and scanned as part of Project Vellum were used to train the Muse Spark family of models,” the complaints say, and that “copyright-related information from these purchased books was not filtered out before being used for model training.” Meta released Muse Spark in April and Muse Glimmer in August, according to the complaints. In answers filed in court on Oct. 6 in a suit brought by Entrepreneur Media, Meta named five people who made or approved decisions on buying or scanning books for Vellum, among them Chris Cox, its chief product officer, and Susan Li, its chief financial officer, whom it identified as having “approved decisions to fund the efforts to physically acquire and scan books as part of Project Vellum.”

The complaints allege only “on information and belief” that the plaintiffs’ own books were among those scanned, an inference they draw from Meta’s buying books “at scale, across every genre of published work.”

The Precedent

In June 2025, two judges of the same court ruled two days apart. Judge William Alsup, in a case brought by authors against Anthropic, held that training on their books was fair use, and that converting print books Anthropic had bought into digital copies was fair use too, for a different reason: each digital copy replaced a print copy, which Anthropic or its vendors discarded. “The print original was destroyed. One replaced the other,” he wrote. ChatGPT Is Eating the World, a blog that tracks AI copyright suits, counts his as the only federal ruling on scanning.

Vellum’s copies, the complaints say, “were not a one-for-one change of format for Defendant’s own reading or reference” but “wholesale reproductions of entire works, made and retained in multiple copies.” “Scanning a book is reproduction,” they say, and they quote Meta as saying that “[a] number of transitory and persistent copies of each of these works would have been created.”

Scanning a book is reproduction.

The suits are likely to go to Judge Vince Chhabria, who presides over the authors’ and publishers’ suits against Meta, the blog says. On June 25, 2025, in Kadrey v. Meta, he granted Meta summary judgment against 13 authors who said that training its Llama models on their books broke copyright law, but wrote that the ruling “does not stand for the proposition that Meta’s use of copyrighted materials to train its language models is lawful.” He also wrote that “the plaintiffs are not entitled to the market for licensing their works as AI training data,” the market that, by the new complaints’ account, Vellum “was designed to bypass.” He called market dilution, a model flooding the market with similar works, “the potentially winning argument”; the new complaints make it. But he also wrote that “fair use doesn’t require that the secondary user make the lowest number of copies possible,” in a ruling on Meta’s earlier training of Llama, not on Vellum.The Other CasesBesides the consolidated Kadrey case, Meta’s July quarterly report lists nine suits with similar claims filed against it since November 2025. As of that report, summary-judgment hearings in Kadrey and in four of the others were set for Feb. 25, 2027.

Meta did not immediately return Bloomberg Law’s request for comment. AI companies, including Meta, have argued that their conduct is fair use because they “transform the copies into an entirely new technology,” as Bloomberg Law put it. In its quarterly report in July, Meta told shareholders that statutory damages for copyright “are calculated on a per work basis, which may result in substantial damages, particularly given the large volumes of data required to train AI models.”METANasdaq614.99▼9.53−1.5%Meta Platforms · Friday’s session

Meta reports third-quarter results after the market closes on Wednesday; the authors’ lawyers, Bloomberg Law reported, plan suits for close to 1,000 more authors in the coming months.