Mark Zuckerberg, the CEO of Meta, delivered a keynote address during the Meta Connect annual event at the company’s headquarters in Menlo Park, California, on September 25, 2024.
Manuel Orbegozo | Reuters
Meta has successfully defended itself in a significant copyright lawsuit from a group of 13 authors regarding the usage of the company’s Llama AI model, although the judge clarified that the ruling is specific to this case.
U.S. District Judge Vince Chhabria endorsed Meta’s stance that the company’s utilization of books for training its large language models, or LLMs, falls under the fair use framework of U.S. copyright law.
The plaintiffs, which include well-known authors like Sarah Silverman and Ta-Nehisi Coates, accused Meta of infringing copyright laws by not obtaining permission to use their books for the AI model, among other accusations.
Chhabria acknowledged that while it is typically unlawful to reproduce protected works without consent, the plaintiffs did not effectively demonstrate that Meta’s use of books for Llama training inflicted “market harm.” He remarked that the arguments presented by the plaintiffs were flawed.
“Meta has successfully countered the plaintiffs’ weak claim that its actions cause or threaten significant market harm,” Chhabria noted. “This conclusion may conflict with reality.”
The judge affirmed that Meta’s “transformative purpose” in copying the works is safeguarded by fair use principles.
A Meta spokesperson commented, “We welcome today’s decision from the Court. Open-source AI models are driving revolutionary advancements, productivity, and creativity for both individuals and businesses, and fair use of copyright materials is essential for developing this groundbreaking technology.”
While there are legitimate concerns that Meta’s data training practices could negatively affect the book market, the judge stated the plaintiffs did not sufficiently argue their case.
Legal representatives for the plaintiffs did not reply to a request for comment.
Moreover, Chhabria identified weaknesses in Meta’s defense, asserting that the public interest would be “seriously undermined” if companies were restricted from using copyrighted material for training purposes without compensation.
“Meta seems to suggest that such a ruling would halt the progress of LLMs and other generative AI technologies, which is absurd,” Chhabria commented.
The judge allowed the possibility for other authors to file similar copyright lawsuits against Meta, remarking that the broader implications of the ruling are minimal.
“This is not a class action, so this ruling only impacts the rights of these thirteen authors — it does not affect numerous others whose works were utilized by Meta for training its models,” he detailed. “And, as is now evident, this ruling does not imply that Meta’s use of copyrighted materials for training its language models is legal.”
Additionally, Chhabria pointed out that a separate claim against Meta is still in progress, regarding allegations that the company may have illegally distributed the plaintiffs’ works via torrenting.
Earlier this week, a different federal judge determined that Anthropic’s use of books for training its AI model Claude was also “transformative,” thereby aligning with the fair use doctrine. However, that judge ordered Anthropic to face a trial concerning claims that it downloaded millions of pirated books for training its AI systems.
“The fact that Anthropic later purchased a copy of a book it previously stole from the internet does not absolve it of liability for the theft, but it might influence the amount of statutory damages,” the judge noted.
WATCH: Meta contests the ban on WhatsApp for devices utilized by House of Representatives staff.
