Anthropic legally purchased the books it used to train its model according to the judge. And the judge said that was fine. Anthropic also downloaded books from a pirate site and the judge said that was bad -- even though the judge also said they didn't use those books for training....

Books.Google.Com was deemed fair use because it only shows previews, not full downloads. Internet Archive is still under litigation iirc besides having owned a physical copy of every book they ever scanned (and keeping a copy in their warehouses) they let people read the whole thing.

I’m surprised Google hasn’t hit its competitors harder with the fact that they actually got permission to scan books from its partner libraries and Facebook and OpenAI just torrented books2/books3, but I guess they have aligned incentive to benefit from a legal framework that doesn’t look to closely at how you went about collecting source material