Meta can no longer reduce its book-torrenting controversy to a handful of employees who happened to run download commands. A federal magistrate judge has ordered the company to identify the people who approved obtaining books from shadow libraries, including LibGen and Anna’s Archive, for artificial-intelligence training. The discovery order puts Meta’s internal chain of command under the microscope as copyright plaintiffs pursue claims over how the material was acquired and shared.
U.S. Magistrate Judge Thomas S. Hixson’s October 19 order requires Meta to provide the names of individuals involved in acquiring the datasets, along with approximate dates and their connection to the decisions at issue. The inquiry applies not only to engineers who performed downloads, but also to supervisors, managers, executives, and groups that authorized the work.
The order specifically addresses Meta’s use of BitTorrent to obtain LibGen Scimag and material from Anna’s Archive. Plaintiffs want to know who decided to use peer-to-peer downloading and who determined whether files should be “seeded”—uploaded in pieces to other users—or whether those uploads should be limited or prevented. Because torrent software can distribute data while downloading it, that distinction could become central to claims that Meta did more than merely copy books for internal training.
Meta has characterized its early dataset-collection process as an informal team effort rather than a series of decisions made by identifiable leaders. The company has also argued that it knows of no individual who chose to distribute the material. Hixson rejected that narrow framing for discovery purposes, finding that plaintiffs are entitled to identify participants even when decisions emerged from a group rather than one named executive.
The dispute spans a class action brought by authors including Richard Kadrey and Sarah Silverman and a separate lawsuit filed by publisher Entrepreneur Media. The cases focus on Meta’s use of books to develop its Llama language models, but the new order does not decide whether Meta infringed copyright or whether any particular executive authorized piracy. It determines what information Meta must disclose as the litigation continues.
That distinction matters because Meta has already won an important ruling in the authors’ case: on June 25, 2025, a judge found that copying the named plaintiffs’ books to train its AI models constituted fair use on the record before the court. However, claims tied to distribution through torrenting remained unresolved. The Copyright Office’s summary of Kadrey v. Meta Platforms likewise describes Meta’s acquisition of shadow-library book datasets through torrenting and their use in training Llama models.
The next phase will turn an abstract corporate-liability fight into a more concrete reconstruction of who knew what, when they knew it, and how Meta’s data pipeline operated. If the disclosures identify senior personnel or show that torrenting automatically exposed files to outside peers, plaintiffs could use that evidence to pursue the unresolved distribution claims—even though the fair-use ruling narrowed the authors’ case over the training itself.
