“Torrenting from a corporate laptop doesn’t feel right”: Meta emails unsealed - Ars Technica

Newly unsealed emails reveal evidence against Meta in a copyright case, alleging the company illegally trained its AI models on pirated books. Meta admitted to torrenting the LibGen dataset, but the emails show the extent of the torrenting, including 81.7 terabytes of data from various shadow libraries.

Internal communications show Meta employees expressing concerns about the legal risks of torrenting and seeding copyrighted material. Despite these warnings, Meta allegedly continued downloading and seeding data, even attempting to conceal their actions. Authors claim Meta's actions contradict previous testimony, including Mark Zuckerberg's claims of non-involvement, and are demanding further depositions.