Meta Ordered to Name Executives Behind AI Training Torrents
Judge orders Meta to name who approved shadow-library torrents and release server logs in copyright cases over its Llama 1 and 2 AI training datasets.
Summary
As of October 20, 2026, Magistrate Judge Thomas Hixon had ordered Meta to identify everyone who made, authorized, or approved obtaining shadow-library datasets, including their involvement dates, decisions and datasets, or swear that nobody qualified. The discovery covers a class action involving authors Richard Kadrey and Sarah Silverman and a separate Entrepreneur Media lawsuit. Meta must also produce command histories for every torrenting server, virtual machine and AWS instance, and explain what happened to copyrighted works not used for AI training.
Meta argued that early dataset acquisition for Llama 1 and 2 was informal and collaborative, with no single authority making formal decisions, and proposed providing names without roles or approval details. Hixon rejected that narrow interpretation. Staff testimony indicated torrenting required director or vice president approval, leadership paused and resumed Anna’s Archive downloads, and LibGen downloading followed a prior escalation to Mark Zuckerberg, identified internally as “MZ.” Meta has claimed potentially identifying communications are privileged.
Plaintiffs say Meta repeatedly torrented shadow libraries from 2022 to 2024, downloading 80.6 terabytes from LibGen and 35.7 terabytes through Anna’s Archive. They argue torrent swarms necessarily facilitated distribution, while Meta says mitigations prevented seeding and that it knows of nobody who approved uploads. The server histories could establish whether Meta seeded copyrighted works, potentially supporting contributory infringement claims that require proof Meta facilitated transfers.
Positives
- Hixon ordered Meta to identify every person involved in approving shadow-library downloads, with dates, decisions and datasets.
- Command histories from every torrenting server, virtual machine and AWS instance could establish whether copyrighted works were seeded.
- Richard Kadrey, Sarah Silverman and other plaintiffs gained access to evidence potentially supporting contributory infringement claims.
- Meta must explain what happened to copyrighted works downloaded but not used for AI training.
Risks & concerns
- Meta downloaded 80.6 terabytes from LibGen and 35.7 terabytes through Anna’s Archive between 2022 and 2024.
- Staff testimony indicated torrenting required director or vice president approval despite Meta’s claim that nobody formally decided.
- Internal messages linked the LibGen decision to a prior escalation to Mark Zuckerberg, identified as “MZ.”
- Server logs could undermine Meta’s claim that mitigation measures prevented seeding copyrighted books.
- Meta maintains that potentially identifying approval communications are privileged, limiting plaintiffs’ access to parts of the decision trail.