Looking at the complaint ( https://publishers.org/wp-content/uploads/2026/05/2026-05-05... ), that seems like the part that's got the most solid foundation, especially given that while torrenting the books, they were also seeding to other peers.

The items they call out around training the models (and attempting to claim that each subsequent model generation should count as an additional instance of infringement) seem far less grounded in the current court interpretations of AI training.