Meta won an important round in a copyright fight over AI training, but the ruling was far narrower than a simple victory headline suggests. Judge Vince Chhabria sided with Meta in a case brought by book authors, while making clear that the decision should not be read as a blanket approval of using copyrighted books to train language models.
The case turns on fair use, market harm, and what plaintiffs must prove when they argue that generative AI threatens the value of their work. It also shows that courts may agree that AI training can be transformative while still disagreeing sharply about how much weight to give the economic impact on authors.
Meta won because the authors missed the strongest argument
Chhabria granted Meta’s request for summary judgment, but he said the reason was the way the plaintiffs built their case. His order emphasized that the ruling did not establish that Meta’s use of copyrighted material to train its language models was lawful in general.
“This ruling does not stand for the proposition that Meta’s use of copyrighted materials to train its language models is lawful,” Chhabria wrote. “It stands only for the proposition that these plaintiffs made the wrong arguments and failed to develop a record in support of the right one.”
According to the ruling, the authors focused on two theories. They argued that users of Llama could reproduce text from their books, and that Meta’s copying harmed a licensing market for copyrighted materials used in AI training.
Chhabria found both theories flawed. The first failed because Llama could not produce long excerpts from the works, even with adversarial prompting. The second failed because, in his view, authors are not entitled to control the market for licensing books for AI training.
The argument Chhabria suggested could have mattered more was different: whether Llama could flood the market with competing AI-generated books and indirectly hurt book sales. The plaintiffs did not introduce evidence showing that kind of market dilution, so Meta was able to rely on expert testimony that Llama 3’s release did not have any discernible effect on the plaintiffs’ sales.
The ruling leaves room for future author lawsuits
Even though Meta prevailed, Chhabria treated the outcome as limited. The decision came before class certification, so it applies only to the 13 authors who sued Meta in this case.
That matters because other authors could try to bring a stronger record. Chhabria wrote that plaintiffs may often win in cases involving uses like Meta’s when they develop better evidence on market effects.
“No matter how transformative [AI] training may be, it’s hard to imagine that it can be fair use to use copyrighted books to develop a tool to make billions or trillions of dollars while enabling the creation of a potentially endless stream of competing works that could significantly harm the market for those books.”
That language is important because it separates two questions. A court may accept that AI training transforms copyrighted material into a different kind of technology, while still finding that the resulting product threatens the market for the original works. For Chhabria, market harm remains central.
The only surviving claim in the case concerns Meta’s controversial torrenting of books to train Llama. The authors have so far successfully alleged that Meta may have violated copyright laws by distributing their works as part of the torrenting process.
Judges are split on the schoolchildren analogy
The Meta ruling arrived soon after a landmark ruling in a separate Anthropic case. In that case, Judge William Alsup deemed Anthropic’s copying of books to train artificial intelligence models a transformative fair use.
Chhabria did not fully accept Alsup’s framing. He criticized the Anthropic ruling for focusing heavily on the transformative nature of generative AI while giving too little weight to potential harm in the markets for the works used in training.
One point of disagreement was Alsup’s comparison between AI training and teaching schoolchildren to write well. Chhabria rejected that analogy when assessing market effects.
“But when it comes to market effects, using books to teach children to write is not remotely like using books to create a product that a single individual could employ to generate countless competing works with a miniscule [sic] fraction of the time and creativity it would otherwise take.”
For Chhabria, the issue is not simply whether learning from books is transformative. It is whether the product built from that copying can create substitutive works at a scale and speed that could damage the market for the originals.
AI companies still face pressure to pay rights holders
Chhabria said his ruling does not create a bright-line rule protecting AI companies. In his view, many circumstances may still make it illegal to copy copyright-protected works to train generative AI models without permission.
That means companies may need to pay copyright holders for the right to use their materials if they want to avoid liability. Chhabria also rejected Meta’s argument that an author victory would stop AI innovation “in its tracks,” calling that position “ridiculous.”
He wrote that if rights holders win lawsuits against AI companies, the result would not necessarily be the end of AI development. Instead, companies could pay authors, rely on materials in the public domain, or prove that copyrighted works are not necessary for AI training.
The ruling may also matter beyond books. Chhabria suggested that news organizations suing OpenAI over allegedly infringing ChatGPT outputs could present stronger arguments against fair use if the outputs indirectly compete with their websites.
Ian Crosby, a lawyer representing The New York Times in that suit, told Ars that both Chhabria’s and Alsup’s rulings are viewed as strengthening the NYT’s case. He said generative AI developers may not build products by copying stolen news content, especially where products output substitutive content that threatens original, human-made journalism.
The main lesson is that AI copyright cases are likely to turn less on broad slogans and more on evidence. Meta won this case because these plaintiffs did not prove the market harm Chhabria considered most important. Future plaintiffs may build that record differently.