Anthropic urged a federal judge in California on Wednesday to rule that its use of song lyrics to train its Claude chatbot is protected as fair use under US copyright law, as the AI company tried to defeat a lawsuit brought by Universal Music Group and fellow music publishers Concord and ABKCO. The hearing before US District Judge Eumi K. Lee in San Jose put the fair use defence — the argument on which dozens of AI training lawsuits now turn — in front of a third Northern California judge, after two of her colleagues reached different conclusions in earlier cases.
The publishers sued Anthropic in 2023, alleging that its use of their lyrics to train Claude infringed copyrights in hundreds of songs by artists including Beyoncé, the Rolling Stones and the Beach Boys. Both sides have asked Lee for summary judgment: the publishers on infringement, Anthropic on fair use and non-infringement. No ruling was issued at the hearing.
What Each Side Argued On Fair Use
Anthropic has argued that its training makes transformative use of song lyrics, in its words to “help Claude understand human language and enable progress and productivity in science, business and education.” Allowing copyright holders to veto such a technology, the company told the court, “would ignore the Supreme Court’s mandate that the primary objective of copyright is not to reward the author, but to serve the public.”
The publishers told the court the use is not fair because it lets Claude generate new lyrics that dilute the market for their works. “Claude is being used to proliferate AI-generated song lyrics in huge numbers, and such AI songs created with tools like Claude are saturating streaming platforms and even climbing the Billboard charts,” the publishers said in a court filing in March. Reporting on Wednesday’s hearing by Bloomberg Law said Lee pressed both sides on exactly that point — the potential harm Anthropic’s copying could cause the publishers, which maps onto the market-effect factor courts weigh in fair use analysis — and on whether this could become the first AI copyright case to reach a trial.
Two Judges, Two Answers Already On The Record
Lee is the third judge in the district to confront the question. Last year, now-retired US District Judge William Alsup sided with Anthropic, holding that its AI training made fair use of books in the separate Bartz case. Two days later, US District Judge Vince Chhabria ruled for Meta in a similar lawsuit over book training but warned that AI training “in many circumstances” would not be fair use.
Lee has also touched this dispute before. In March 2025 she refused to block Anthropic from using the lyrics for training at the preliminary stage, saying the publishers were asking her to define a licensing market while the fair use question remained unsettled. Wednesday’s hearing was different in kind: cross-motions for summary judgment ask the court to decide the fair use defence on the merits, not to preserve the status quo.
Why It Matters
The music publishers’ case matters because lyrics are a harder test for the training defence than books were. A song lyric is short, highly creative and licensed through a mature, centralised market — the exact combination the market-harm factor is designed to catch. If Lee finds fair use here, the reasoning would extend well beyond music to news text, images and other short-form works that AI companies train on at scale. If she finds against Anthropic, or sends the case to a jury as the first AI copyright trial, every lab training on licensed creative catalogues faces a priced-in licensing obligation it does not currently carry. Anthropic’s position is sharpened by its split record: the company won the books fight before Alsup, and a loss on lyrics would draw a line between categories of training data rather than between companies. The ruling will also land while the wider set of copyright fights reshaping AI — news publishers, authors and image libraries among them — is still moving through the same courts, and judges in those cases are reading each other’s fair use reasoning closely.
Background
The publishers filed the case in 2023, in the first wave of lawsuits over generative AI training data, after earlier disputes over Claude reproducing song lyrics in its answers. The litigation has since consolidated into one of the reference cases for AI training law, alongside the books cases against Anthropic and Meta and the news cases against OpenAI. Anthropic, founded in 2021 by former OpenAI researchers, builds the Claude family of models and has made its legal posture on training data — including a settlement in the books case — part of its public identity. Related coverage: how Anthropic and OpenAI differ on safety and strategy and what Anthropic is, explained.
Source: Reuters, Bloomberg Law
0 Comments