Meta’s win came in a lawsuit brought by a group of 13 authors, including comedian Sarah Silverman. The authors claimed Meta used their books without consent to train its Llama AI model. However, Judge Vince Chhabria of the Northern District of California ruled that the authors failed to demonstrate concrete harm to the market value of their books. He added that the use of their content was “highly transformative” and served a different purpose than the original works.
“The plaintiffs failed to show how Meta’s AI models serve as substitutes for their books,” Judge Chhabria stated. “The models do not output anything close to the expressive style or content of the original works.”
Anthropic’s case offered a more nuanced outcome. Judge William Orrick ruled in favor of Anthropic on the core issue of using copyrighted texts in training, calling the process “quintessentially transformative.” However, he also found that Anthropic had stored more than 7 million pirated books in what was essentially a “central library” used in training its Claude models—an act that could constitute copyright infringement.
That part of the case will now proceed to trial in December 2025, where Anthropic could face significant damages if found liable.
Despite these victories, both rulings come with clear warnings. The judges emphasized that these decisions are narrowly tailored to the facts presented and do not establish blanket legal protections for AI developers using copyrighted materials. “This ruling should not be read as giving carte blanche to all uses of copyrighted works in AI training,” noted Judge Orrick.
Meanwhile, lawsuits from The New York Times against OpenAI and Microsoft, and from music publishers against AI firms, are still pending, and could bring very different outcomes based on the type of content used and potential market impact.
