Meta is facing mounting pressure in a consolidated series of copyright lawsuits in California federal court, as new deposition testimony and internal documents complicate the tech giant’s defense regarding its use of pirated books to train its Llama artificial intelligence models. The legal battles, overseen by U.S. District Judge Vince Chhabria in the U.S. District Court for the Northern District of California, bring together a high-profile roster of plaintiffs. Among them are celebrated authors such as Sarah Silverman and Ta-Nehisi Coates, alongside prominent publishing entities including Entrepreneur Media and Chicken Soup for the Soul.
The core of the legal dispute revolves around the controversial Books3 dataset—a massive compilation of nearly 200,000 books extracted from the library of Bibliotik, a private torrent tracker. Back in August, Meta informed Entrepreneur Media that it did not believe it had acquired Books3 via BitTorrent. However, recent developments in the discovery phase have cast significant doubt on that assertion, pulling former employees and internal corporate communications directly into the crosshairs of the litigation.
Books3 Torrenting Testimony
According to a joint status report filed by the rightsholders, explosive deposition testimony from former Meta researcher Stephen Roller paints a very different picture. Roller, who worked at Meta within its Fundamental AI Research (FAIR) group between 2017 and 2023, reportedly admitted during a deposition that he torrented Books3 and knowingly distributed it during his tenure at the company.
While Roller could not recall the exact date he acquired the dataset, plaintiffs noted that he acknowledged Books3 was created in 2020 and that he maintained no other concurrent employment during his time at Meta. This timing is critical to the progression of the case. Last year, the court ruled that Meta’s training of the Llama models on books constituted fair use under copyright law. Consequently, the method by which Meta acquired its copies—specifically via BitTorrent downloads—has emerged as the primary open issue remaining in the authors’ case. Earlier this month, a judge ordered Meta to identify the specific employees who approved downloading materials from shadow libraries.

Meta has previously acknowledged utilizing Books3. In early 2024, the company admitted that it had used portions of the dataset to train its Llama 1 and Llama 2 large language models, though that initial admission remained silent on how the corporation actually obtained the data. Plaintiffs now argue that Roller’s testimony goes much further. Roller reportedly stated that Meta held copies of Books3 by the time he departed the AI research group, and he explicitly named two other research scientists who likely acquired their own copies.
Interestingly, Roller no longer possesses his own copy of the dataset. Plaintiffs state that he deleted it approximately a year ago, likely after receiving a second subpoena. He reportedly purged the files "due to all this litigation," feeling it "seem[ed] contentious to have copies of The Pile," a broader text dataset that incorporates Books3.
Meta: Could Have Been for Himself
Meta has sharply pushed back against the plaintiffs’ interpretation of the deposition, arguing in the joint filing that the rightsholders are mischaracterizing Roller’s statements. According to the tech company, this is the first time the plaintiffs have raised these specific concerns, and Roller’s testimony provides no concrete foundation for them.
"Mr. Roller testified that he did not know when he downloaded a copy of the ‘The Pile’ via torrent, and did not testify that his download of the ‘The Pile’ was for Meta or at Meta, noting it could have been for himself or for another employer," Meta writes in its defense.

Furthermore, Meta emphasizes that Roller stated he did not know whether multiple copies were downloaded or whether the two other researchers he mentioned ever actually acquired Books3. Roller left Meta in January 2023 and has subsequently worked for three other artificial intelligence firms. Addressing the deletion of the files, Meta clarifies that the copy was stored on Roller’s personal computer and was removed more than two years after he left the company simply because he "didn’t need it, so [he’d] rather not have it."
A (Pirated) Book Club
The legal friction is not restricted to AI training data. The joint status report highlights other instances where pirated literature allegedly circulated within Meta, raising questions about whether unauthorized texts were utilized for purposes outside of machine learning development.
Plaintiffs are now pushing to compel Meta to produce the contents of an internal Google Drive folder labeled "ML Books," alongside records from an internal Meta book club that allegedly relied on materials sourced from shadow libraries. Discovery documents previously produced by Meta revealed an internal exchange where an employee shared a "pirated book" with a colleague. Recognizing the irony given the company’s legal battles, the employee wrote, "i know for vellum sake this is prob very ironic." When pressed by her colleague about whether the text was indeed pirated, she allegedly responded by sharing a link to the "ML Books" folder.
The recipient of the link expressed immediate concern over the corporate liability, messaging another coworker, "bro she tryna get me fired," and adding that he "may need her to delete those messages."

Plaintiffs contend that this internal chatter is highly relevant. If Meta employees were routinely utilizing shadow library books for informal corporate activities like book clubs, it could further undermine the company’s fair use defense by demonstrating a pervasive internal culture of disregarding copyright restrictions.
Meta, however, argues that these latest demands are untimely. The company notes that the plaintiffs first requested access to the linked files on October 9—more than a month after the formal discovery period had already closed. Regarding the "ML Books" folder itself, Meta stated that it is actively investigating the matter and intends to discuss it with the plaintiffs as more information becomes available.
Missing Server Logs
Beyond the disputed provenance of Roller’s files, the plaintiffs maintain that critical electronic evidence remains unaccounted for. Despite prior court orders, Meta has allegedly failed to hand over all command history files associated with its torrenting activities. According to the rightsholders, these technical logs are essential because they illustrate how Meta configured its BitTorrent software, which specific torrent files were targeted, and ultimately which copyrighted works were downloaded.
Plaintiffs assert that Meta has yet to provide command history files from any of the specific servers utilized during the periods when torrenting took place. Consequently, they are asking the court to order Meta to formally state whether any of these command history logs were intentionally or accidentally deleted, and if so, when the deletion occurred.

Meta counters that it has already produced substantial volumes of server logs and continues to comb through archival backups, supplying new files to the plaintiffs on a rolling basis. The corporation maintains there is no evidence to suggest that any command history files were ever deleted.
The broader litigation continues to expand at a rapid pace. Entrepreneur Media’s individual lawsuit is currently scheduled to proceed to trial in the spring of 2027, with lawsuits from other participating publishers lined up sequentially should a settlement be reached beforehand. Meanwhile, the queue of related legal actions continues to grow. Earlier this month, a new wave of major publishers—including Elsevier, Hachette, and Macmillan—along with a growing roster of individual authors, officially joined the legal queue against the tech giant.