In brief
- A U.S. judge ruled Meta’s AI training on copyrighted books qualifies as fair use, dealing a blow to 13 authors.
- The decision follows a similar ruling favoring Anthropic, though courts warned AI training practices remain legally unsettled.
- Chhabria said Meta prevailed only because the authors failed to present strong arguments and evidence.
A federal judge delivered a significant blow to authors suing tech giants over AI training this week. The judge ruled that Meta’s use of copyrighted books to train its artificial intelligence models constituted fair use under copyright law.
U.S. District Judge Vince Chhabria in San Francisco sided with Meta Platforms on Wednesday in a case brought by 13 authors, including comedian Sarah Silverman and Pulitzer Prize winners Junot Díaz and Andrew Sean Greer.
The 13 authors suing Meta failed to provide enough evidence that the company’s AI would dilute the market for their work, Judge Chhabria said in the ruling.
Their argument, he said, “barely gives this issue lip service” and lacked the facts needed to prove harm under U.S. copyright law.
But the judge made clear that the ruling is far from a blanket endorsement of AI companies’ controversial training practices.
“This ruling does not stand for the proposition that Meta’s use of copyrighted materials to train its language models is lawful,” Chhabria said. “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.”
Kunal Anand, CEO of AI chatbot service AiBaat, told Decrypt that he hopes it’s a sign that courts will find a way of “balancing technological progress with creator rights.”
“While the decision favored Meta, it reminds us that ethical AI development demands clear licensing frameworks,” he added.
The authors sued Meta and OpenAI in 2023, alleging the companies misused pirated versions of their books to train their Llama AI and ChatGPT systems without permission or compensation.
In January, court filings revealed that Meta CEO Mark Zuckerberg personally approved using the pirated dataset, despite warnings from his AI team that it was illegally obtained. Internal messages cited in the filings show engineers at Meta hesitated, with one employee admitting, “torrenting from a corporate laptop doesn’t feel right.”
But the company proceeded anyway.
Judge Chhabria acknowledged the potential for AI to “flood the market with endless amounts of images, songs, articles, books, and more” using “a tiny fraction of the time and creativity that would otherwise be required.”
He noted in the ruling this could “dramatically undermine the market for those works, and thus dramatically undermine the incentive for human beings to create things the old-fashioned way.”
Chhabria expressed sympathy for authors’ concerns, but it wasn’t enough to make a sound legal argument. “Courts can’t decide cases based on general understandings,” he said.
The ruling favoring Meta affects only these 13 specific authors since it was not certified as a class action.
The decision marks the second major victory for AI companies this week, following a similar ruling favoring Anthropic on Monday.
In that case, Judge William Alsup also found AI training to be fair use, but criticized Anthropic for building a permanent library of pirated books.
Experts say the solution to disputes over AI training and copyrighted content lies in proactive market-based approaches rather than waiting for regulatory clarity.
“By the time policymakers catch up to the latest AI breakthroughs, those breakthroughs will have advanced another generation,” Hitesh Bhardwaj, co-founder at Capx AI, told Decrypt. “A more sustainable path is to reward people whose work fuels AI: create transparent marketplaces where authors and creators license their own data on fair terms.”
“That approach puts control back in the hands of the people whose content powers our models,” he said.
Edited by Stacy Elliott.
Generally Intelligent Newsletter
A weekly AI journey narrated by Gen, a generative AI model.
