> Since whole chunks of these books can be recited verbatim by these models, to which they sell access, they absolutely are republishing and reselling these books' content in a way.

They are absolutely not "republishing" in any meaningful sense of the term. A chunk is not a whole book, and even getting a modern LLM to reproduce such a large chunk of an arbitrary book is not a trivial task. I have never heard of anyone who actively used LLMs for book piracy.

> Like I remember a research paper that managed to recreate the whole of a Harry Potter book from a model?

Even if that is true (it may well be false), this is likely far too difficult for any normal person to exploit, and moreover, even less likely to succeed for the great majority of other books who aren't nearly as famous.