Meta won a significant victory in a copyright suit on Wednesday, when a federal court ruled that it did not break the law by training its AI tools without permission on the books of 13 authors.
The Court had no choice but grant summary judgment to Meta in the case of the plaintiffs who claimed that Meta violated copyright laws by training its models on their books, wrote US District Court Judge Vince Chhabria. He concluded that plaintiffs had not presented enough evidence to prove that Meta’s use their books was harmful.
A group of high-profile authors, including comedian Sarah Silverman, and journalist Ta-Nehisi coates, sued Meta in 2023, claiming that the tech giant had violated their copyrights by training its large-language models on their works. Kadrey V. Meta is one of the first cases in this area. Now, there are dozens similar AI copyright suits winding their way through US courts. Chhabria had stated in the past that he would carefully examine whether the plaintiffs could prove that Meta’s use their work would harm them financially. In virtually all cases where a defendant has stolen someone’s work without permission, the key question is whether allowing others to copy it would significantly reduce the market for the original. This was what he wrote on Wednesday in his judgment.
The second major ruling on AI copyright this week is that US District Court Judge William Alsup ruled Monday that Anthropic was allowed to use copyrighted material to train its AI tools. Alsup’s ruling gave Anthropic a victory, but it was a split verdict, as the AI firm will still have the plaintiffs in court to answer for pirating the books. Chhabria, unlike Alsup, did not focus his attention on the plaintiffs’ claim in Kadrey that Meta’s use pirated materials was an issue. Instead, he noted that the parties will have a Zoom meeting to discuss how to deal with the piracy allegations. Chhabria’s stance was further differentiated from Alsup by Chhabria stating that Alsup “brushed aside” the importance market harm when he made his fair-use decision by focusing on the use of the “work”. They also look at whether the new work is “harmful” to the original rights holders financially. James Grimmelmann is a professor at Cornell University who specializes in digital and internet law. He says that it’s noteworthy that he disagreed with Judge Alsup, but did so respectfully.
Legal experts also highlighted Chhabria’s focus on market damage, noting that this could shape future AI copyright arguments. Jacob Noti Victor, Cardozo Law Professor, says that we haven’t heard the last of this novel theory on market dilution. “That could change the game for the other cases or future litigation.”
Advocates of the idea that AI is transformative still view Chhabria’s ruling as a victory. Adam Eisgrau, senior director of AI, Creativity, and Copyright Policy for the tech trade group Chamber of Progress, says that Judge Chhabria’s ruling today is a win for advocates of the idea that AI training is transformative. “He didn’t want to come to that conclusion, for reasons he explains and which are out of line with established fair-use precedents, in terms of market harm. Market dilution, however, is a load of nonsense.
That’s the problem. Chhabria was careful to emphasize that his ruling in this case was based on a specific set of facts, leaving the door open for future authors to sue Meta over copyright infringement. “In many situations it will be illegal for copyright-protected work to train generative AI model without permission,” wrote Chhabria. “This means that companies, in order to avoid liability for infringement of copyright, will need to pay copyright holders to use their materials.” “However, the Court does take very serious the idea that AI model trained on plaintiffs books could ‘flood’ the market with endless images, songs and articles, as well as books.’ This would harm the market for original works. He probably takes this more seriously than plaintiffs, since they didn’t provide any evidence. I have never seen a case where a judge has lamented the plaintiffs for not arguing their case as much as this one.
The plaintiffs’ lawyers at Boies Schiller Flexner stated in a press release that AI companies who ‘feed copyright protected works into their models’ without getting permission from copyright holders or by paying for them are generally breaking the law. “Yet despite Meta’s undisputed history of historically unprecedented pirating copyrighted work, the court ruled Meta’s in favor.” We respectfully disagree.
Meta had a more positive response. Thomas Richards, Meta spokesperson, said in a press release that they “appreciate today’s Court decision.” “Open-source AI is powering transformative innovation, productivity, creativity, and for individuals and businesses, and fair usage of copyright material, is a vital framework for building this transformative tech.”
Plaintiffs are closely watching the outcome in other AI cases. Mary Rasenberger is the CEO of the Author’s Guild which is suing OpenAI for copyright infringement. She says that the decision was disappointing, but not in full.
In the grand scheme of the things, this ruling has limited consequences. Chhabria wrote that since this is not a case of class action, the ruling only affects these 13 authors’ rights and not the rights of the countless other authors whose works Meta has used to train its language models. “As should be clear now, this ruling does NOT support the proposition that Meta is using copyrighted material to train its language model.
