Anthropic's early lawsuit win pushes courts forward on fair use
The judge's ruling shows the complexity of bringing a lawsuit against model providers and the need for clarity on fair use in the age of generative AI.
Listen to this article. This audio was generated by AI.
AI lawsuits filed against large language model providers such as Anthropic and OpenAI are starting to see movement, which some believe means that the courts are edging toward clarity on what constitutes fair use when training generative AI models.
On March 25, a California federal judge rejected a bid from a group of music publishers, including Universal Music Group, that sued Anthropic in 2023. In the injunction, UMG and plaintiffs wanted to prevent Anthropic from using any of their music to train its AI-powered chatbot Claude while the lawsuit was ongoing. The judge said the request was too broad.
"Although Publishers have tried to clarify the scope of the proposed injunction, the details remain elusive and poorly defined," U.S. District Judge Eumi K. Lee of the Northern District of California wrote in her ruling. "The complaint incorporates and references 500 songs listed in Exhibit A, but this list is 'illustrative and non-exhaustive.'"
The music publishers initially accused Anthropic of infringing on the copyright by using lyrics from about 500 of their songs to train its model. They said Claude was able to generate exact copies of the lyrics to songs like "Uptown Funk" by Bruno Mars and "Halo" by Beyoncé.
A broad bid
The lack of specificity in the injunctive relief likely made the bid something the court couldn't enforce, according to Michael Bennett, associate vice chancellor for data science and AI strategy at the University of Illinois Chicago.
If you are trying to convince a court to give you an injunctive relief against an Anthropic or OpenAI ... then your request ought to be as specific as possible.
Michael BennettAssociate vice chancellor for data science and AI strategy, University of Illinois Chicago
The plaintiffs' request was also too vague on whether UMG and the other music publishers wanted Anthropic to retrain its models without their music or which songs they didn't want it to train on.
"An injunction requiring Anthropic to retrain already-released models, or to rebuild the training corpus for models in development, could impose unforeseeable costs on Anthropic," according to Lee's ruling.
While this ruling is specific to Anthropic, it provides some insight into what judges expect from those bringing copyright cases against AI model providers.
"If you are trying to convince a court to give you an injunctive relief against an Anthropic or OpenAI because you're convinced that they use your copyrighted material to train their model, then your request ought to be as specific as possible," Bennett said. "If it's not a specific one, if it's vague, then it's going to be denied."
A need for clarity on fair use
While the ruling is an early decision in the court proceedings and did not favor UMG and the other music publishers, it was important for Anthropic.
"If Anthropic lost at this stage, it would have meant that there would be an almost insurmountable burden for it to overcome later in the litigation to reverse the effects of this injunction," said Aaron Rubin, a partner at law firm Gunderson Dettmer. "This now advances things to a more substantive phase where we'll see the fair use argument play out."
The judge's ruling also suggested that the fair use argument needed to unfold before such an injunction could be met, Bennett said.
"The court was indicating that the fair use question is more fundamental ... in that it also very likely needs to be answered before an injunctive relief argument or request can be victorious in this kind of situation," he said.
The AI lawsuit against Anthropic wasn't the only one to see movement this week.
On March 26, a federal judge in New York rejected OpenAI's request to dismiss the copyright infringement lawsuit it is facing against The New York Times. U.S. District Judge Sidney Stein of the Southern District of New York dismissed some claims, but also cleared a path for the case to proceed to trial. The suit is being closely watched because the AI vendor is facing a major publisher with a successful subscription-based business model.
The decisions on both lawsuits and a ruling made by a federal appeals court in Washington, D.C., that generated content without human input is not copyrightable could mean that a ruling regarding fair use is coming soon.
"I would gamble that before we get to the end of January 2026, we see something significant on the fair use side," Bennett said.
Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence software and systems.
