A U.S. federal judge in San Francisco has sided with Meta Platforms, dismissing a copyright infringement lawsuit filed by a group of authors who claimed the company used their books without permission to train its Llama generative AI. The court ruled that Meta's use of copyrighted books for AI training qualifies as fair use under copyright law, marking a significant legal victory for the tech giant.

The lawsuit, initiated in 2023, involved 13 writers, including comedian Sarah Silverman and authors Junot Diaz and Laura Lippman. The authors alleged that Meta used pirated versions of their books to train Llama, infringing on their copyrights and allowing Meta to profit unfairly from their creative work. They argued that Llama could even reproduce parts of their works when prompted.

Judge Vince Chhabria acknowledged that unauthorized AI training could be illegal but stated that the authors failed to provide sufficient evidence that Meta's conduct had harmed or would harm the market for their work. He also noted that the plaintiffs "made the wrong arguments". Despite this, Chhabria suggested that future claims with stronger arguments could be more successful. He warned that using copyrighted books to build AI tools capable of flooding the market with competing works might not always be protected by fair use, especially when those tools generate vast profits.

Meta defended its approach, arguing that open-source AI drives innovation and relies on fair use as a key legal principle. The company stated that copyright law allows the unauthorized copying of a work to transform it into something new and that the AI-generated expression from its chatbots is fundamentally different from the books it was trained on. Meta also said it appreciates the decision.

The ruling has sparked debate within the legal and AI communities. Lawyers for the authors expressed their disagreement, stating that AI companies that use copyright-protected works without permission are generally violating the law. They argued that Meta should have paid to license the literary works. However, Judge Chhabria also said that the ruling is limited to the authors in the case and does not mean that Meta's use of copyrighted materials is lawful.

The decision follows a similar ruling in a case against Anthropic, where a judge initially ruled that training AI on copyrighted materials constitutes fair use. However, that judge also ordered Anthropic to answer for using materials from online libraries known to offer pirated books. These back-to-back rulings represent the first federal rulings to address AI training under U.S. copyright law, signaling that courts remain divided on the issue.

The Meta case underscores the complex legal questions surrounding AI development and copyright law. While this ruling is a win for Meta, Judge Chhabria's reservations suggest that the legal landscape is still evolving and that future cases could have different outcomes. The ruling does not confirm the legality of Meta's actions, only that the plaintiffs made weak arguments. It remains to be seen how this decision will impact the future of AI training and the rights of copyright holders.