Judge orders Meta to name execs who approved AI book torrents
A judge has ordered Meta to reveal which executives authorized pirating shadow libraries for AI training, undermining the tech giant's defense that junior employees acted without corporate approval.
Magistrate Judge Thomas Hixon ruled that Meta must identify every individual who authorized or approved downloading datasets from shadow libraries like LibGen and Anna’s Archive between 2022 and 2024. The ruling stems from consolidated copyright lawsuits brought by authors including Richard Kadrey and Sarah Silverman, alongside publisher Entrepreneur Media. The plaintiffs allege Meta illegally distributed copyrighted material via BitTorrent to amass massive training corpora, including 80.6 terabytes of data from LibGen and 35.7 terabytes from Anna’s Archive.
Meta attempted to shield leadership by claiming its early AI training efforts for models like Llama 1 and Llama 2 relied on an informal process where no high-level personnel made explicit decisions. However, court filings show employees expressed discomfort with the practice, with one worker noting that "torrenting from a corporate laptop doesn't feel right." Unsealed internal communications indicate at least one decision to download from LibGen followed an escalation directly to Mark Zuckerberg, contradicting Meta's claim that no formal corporate authorization took place.
As part of the discovery order, Judge Hixon also compelled Meta to release command history logs for all servers, virtual machines, and Amazon Web Services (AWS) instances utilized during the torrenting operations. Plaintiffs intend to use these server logs to determine whether Meta seeded copyrighted files to other swarm peers or deployed mitigations to prevent outbound sharing.
For enterprise AI developers and practitioners, this court decision signals an end to organizational ambiguity surrounding data sourcing. Tech companies can no longer rely on diffuse internal workflows to avoid corporate liability for copyright infringement during pre-training. Moving forward, AI engineering teams will likely face strict governance protocols, required paper trails, and explicit legal oversight before integrating external datasets into large language model pipelines.
This is our own summary of reporting by Ars Technica AI
