Bartz v. Anthropic
The US authors' lawsuit that drew the line between training on copyrighted books, which a judge allowed, and pirating them, which he did not.
Bartz v. Anthropic is the case a group of authors brought against Anthropic over books used to train Claude. In 2025 the judge split the question in two, and that split has shaped nearly every AI copyright case since. Using a lawfully obtained copyrighted book to train a model, he held, can be fair use, because the model is not a substitute for the book. Downloading that book from a pirate library to get it in the first place is not.
Anthropic settled, and the 1.5 billion dollar agreement was approved in July 2026. The practical consequence is that the fight has largely moved off the philosophical question of whether training is allowed and onto the far more concrete question of where the files came from, which is why later complaints spend so much space on torrenting.