Publishers Sue Meta and Zuckerberg: They Torrented Millions of Pirated Books to Train Llama — and Masked Their IPs

Meta Torrented Pirated Books to Build Its AI — and Hid Its Tracks

For years, AI companies insisted their training data was a matter of 'publicly available' material and 'fair use.' A lawsuit filed May 5, 2026 alleges Meta's process was considerably less polite: it torrented millions of copyrighted books and journal articles from pirate sites, then took deliberate steps to avoid getting caught.

The Plaintiffs

The suit, filed in the U.S. District Court for the Southern District of New York, was brought by five of the largest publishing houses in the world — Elsevier, Cengage, Hachette Book Group, Macmillan Publishers, and McGraw Hill — alongside bestselling author Scott Turow. Critically, it names not just Meta Platforms Inc. but CEO Mark Zuckerberg personally as a defendant.

What They Allege

The complaint accuses Meta of willful copyright infringement: torrenting millions of copyrighted works from pirate sites to train its Llama large language models. It goes further, alleging Meta 'masked its IP addresses to avoid detection' — an active effort to conceal the piracy — and 'stripped CMI, including copyright notices, from the works.' Removing copyright-management information is itself a separate violation under U.S. law.

Why This One Bites

Earlier AI copyright cases leaned on individual authors. This one is led by institutional publishers with established licensing infrastructure and documented market harm — plaintiffs positioned to show exactly what the legitimate market for that material was, and that Meta chose to torrent it instead. The masking of IP addresses is the detail that converts a fair-use debate into something that looks like a knowing scheme.