Meta Secretly Trained Its AI on a Notorious Piracy Database, Newly Unredacted Court Docs Reveal

Key Takeaways

• Meta lost a legal battle regarding its AI training practices, with a court unsealing documents revealing the company used pirated books from Library Genesis (LibGen) to train its AI models.
• The lawsuit, initiated by authors alleging copyright infringement, challenges the legality of using copyrighted material for AI training.
• The court criticized Meta for attempting to redact information to avoid negative publicity, not to protect business interests, and ordered the release of the unredacted documents.
• These documents include internal communications showing Meta employees were aware of LibGen's pirated nature and that the use of this data was escalated to CEO Mark Zuckerberg.
• The plaintiffs argue that Meta not only used copyrighted material without permission but also distributed it via torrenting, while Meta claims the plaintiffs knew about the LibGen use earlier.