Magistrate Judge Thomas Hixon issued the discovery order in a class action by authors including Richard Kadrey and Sarah Silverman, and in a separate suit by publisher Entrepreneur Media.
Meta has spent weeks pointing to staff who downloaded torrents, arguing that no one with authority "decided" to pirate books.
Meta has said its early AI data collection, including for the Llama 1 and 2 models, ran on an "informal approach" without formal decisions, and that structured approval processes came later.
Plaintiffs counter with staff testimony and messages: one worker said torrenting from a corporate laptop "doesn't feel right," another said downloads needed director- or VP-level approval, and a third said leadership halted and then resumed downloads from Anna's Archive.
Unsealed messages also allegedly show the LibGen decision followed an escalation to Mark Zuckerberg.
Hixon rejected Meta's narrow readings, noting the request covers all decision-makers, not a single one or only formal decisions.
Hixon also ordered Meta to produce command history files for every server it used to torrent, including virtual machines and AWS instances. Authors say these logs could show whether Meta uploaded, or "seeded," their works, which they argue was unavoidable given how torrenting works and the 80.6 terabytes taken from LibGen and 35.7 terabytes from Anna's Archive. Meta says it used "mitigations" to prevent seeding and that it knows of no one who approved uploading. Under a contributory infringement claim, authors need only show Meta facilitated transfers of their works. Meta must also answer questions about what happened to pirated works not used for AI training.