Meta just notched a major courtroom win in the generative AI copyright wars. A federal judge dismissed claims from 13 authors — including comedian Sarah Silverman — that Meta illegally used their books to train its Llama AI models, ruling the company’s actions fell under the “fair use” exception.
Key Points:
- Judge ruled Meta’s AI training on copyrighted books qualifies as fair use.
- Authors failed to prove Meta’s models harmed the market for their work.
- The decision echoes a similar ruling in favor of Anthropic earlier this week.
This case, Kadrey v. Meta, is the first major AI copyright suit to reach a summary judgment — and it’s a big one. Judge Vince Chhabria didn’t just side with Meta; he shredded the authors’ arguments, calling them “clear losers” on the two claims they put forward: that Meta’s models can regurgitate portions of their books, and that their ability to license content for training was harmed. Neither stood up to scrutiny.
Meta admitted it downloaded the books from shadow libraries like LibGen and used them to train Llama 1 and 2. But the court found that Meta’s use was “highly transformative,” since the purpose of training a general-purpose large language model is radically different from simply reading a book for entertainment or education. While Chhabria acknowledged that AI training could, in theory, damage markets for human-created work — especially for lesser-known authors — he said the plaintiffs failed to present “meaningful evidence” that Meta’s models are doing that.
That failure mattered. As the judge emphasized, even transformative uses can be illegal if they undercut the market for the original work. But these authors didn’t even try to make that argument until late in the process, and when they did, it was vague and unsupported. “Speculation is insufficient,” Chhabria wrote.
Notably, the court didn’t bless AI training wholesale. Chhabria made clear that in many cases, using copyrighted works without permission to train generative AI is likely illegal — and that companies like Meta will “generally need to pay copyright holders.” But this case, he wrote, only shows that “these plaintiffs made the wrong arguments and failed to develop a record in support of the right one”.
The decision marks the second major win for the AI industry this week, following Anthropic's victory on Monday in a similar lawsuit over its Claude AI system. Together, these rulings are establishing crucial legal precedent as the tech industry faces over thirty copyright infringement lawsuits by copyright owners against AI developers.
The Meta ruling only affects these 13 specific authors since it wasn't a class action, leaving "countless others whose works Meta used to train its models" free to sue separately. And crucially, Judge Chhabria emphasized his ruling "does not stand for the proposition that Meta's use of copyrighted materials to train its language models is lawful"—just that these particular plaintiffs made weak arguments.
The legal landscape is still forming. The U.S. Copyright Office recently released a report that was more sympathetic to copyright holders, endorsing a new "market dilution" theory that no court has yet adopted. Meanwhile, The New York Times case against OpenAI continues to advance, with a judge recently allowing broader claims to proceed.
The one unresolved piece: Meta’s alleged re-uploading of pirated books during torrenting. That distribution claim will move forward, though it’s a much narrower issue. For now, Meta — and Big AI more broadly — can breathe a little easier.