A federal judge in California has ruled in favor of Meta in a closely watched copyright lawsuit, dismissing claims brought by authors including Sarah Silverman and Junot Díaz over the company’s use of pirated books to train its AI model, Llama. But while the decision represents a significant legal win for Meta, the ruling stopped short of affirming the legality of using copyrighted materials for AI training, and instead faulted the plaintiffs for failing to make their case.

In a detailed opinion, U.S. District Judge Vince Chhabria said the authors failed to provide empirical evidence that Meta’s use of their books harmed the market for their work. “All the plaintiffs presented is speculation,” Chhabria wrote. “Speculation is insufficient to raise a genuine issue of fact and defeat summary judgment.” He noted that a potentially stronger argument — that generative AI could depress the market for human-created works — was barely mentioned by the plaintiffs.

Meta allegedly trained its Llama models in part using LibGen, a so-called “shadow library” that hosts millions of pirated books and academic works. While acknowledging that this could present legal concerns, the judge emphasized that the current ruling does not establish that Meta’s conduct is lawful — only that the plaintiffs failed to support their claims effectively. “This ruling does not stand for the proposition that Meta’s use of copyrighted materials to train its language models is lawful,” Chhabria clarified.

Despite granting summary judgment to Meta, Chhabria warned of a future in which AI firms may be required to compensate copyright holders. “In many circumstances it will be illegal to copy copyright-protected works to train generative AI models without permission,” he wrote. “Companies will generally need to pay copyright holders for the right to use their materials.”

The ruling follows a similar decision earlier this week in favor of AI startup Anthropic. In that case, Judge William Alsup found that the company’s use of copyrighted materials to train its Claude model was “exceedingly transformative” and qualified as fair use. While both decisions offer near-term victories for AI developers, they underscore the legal uncertainty still surrounding generative AI training.

Meta welcomed the decision, describing fair use as a “vital legal framework” for developing transformative AI systems. The plaintiffs’ legal team at Boies Schiller Flexner, however, issued a statement disagreeing with the outcome and accusing Meta of engaging in “historically unprecedented pirating” of copyrighted works.

Chhabria has yet to rule on a separate claim regarding Meta’s alleged distribution of copyrighted material through file-sharing processes.

TLDR:

A federal judge dismissed a copyright lawsuit against Meta over its AI model training, not by affirming its legality, but by finding the plaintiffs failed to make their case. The ruling warns that future AI training on copyrighted content could still be illegal unless creators are compensated.