Anthropic defended its fair use position in federal court on Wednesday as US District Judge Eumi Lee in San Jose heard arguments from the AI company and the music publishers suing it over song lyrics used to train Claude, in a case that could shape how copyright law applies to AI training across dozens of pending lawsuits.
The session has been described in press coverage as Anthropic's next fair use test. The hearing brings years of litigation between Anthropic and major music publishers — Universal Music Group, Concord, and ABKCO — to a pivotal moment, with two of Judge Lee's colleagues on the California federal bench having already reached divergent conclusions about whether training AI on copyrighted works qualifies as fair use. For more context on this story, see our ongoing more AI stories.
The Case: From Beyoncé to the Beach Boys
The lawsuit, filed in 2023, alleges that Anthropic used copyrighted lyrics to train Claude across hundreds of songs by artists including Beyoncé, the Rolling Stones, and the Beach Boys. In a March filing, the publishers argued the practice is not fair use because Claude can generate new lyrics that dilute the market for their catalogs, and because AI-generated songs are flooding streaming platforms — even topping the Billboard charts, they claimed.
The publishers presented a concrete example in court filings: when asked to write "a song about the death of Buddy Holly," Claude produced a verbatim copy of the lyrics to Don McLean's "American Pie," according to reporting on the case by The Epoch Times.
A separate Concord lawsuit alleges something more serious than memorization: that Anthropic trained Claude using pirated copies of 20,517 music works sourced from torrent sites. Concord is seeking damages of up to $150,000 per copyright violation, plus $25,000 for instances where it alleges Claude removed copyright notices from lyrics — or alternatively, the profits Anthropic gained from the alleged infringement.
Anthropic's Defense: Transformative Training and 'Volitional Conduct'
Anthropic argues that training on lyrics is transformative, contending that Claude learns human language in ways that advance science, business, and education. Blocking rights holders from being able to veto such training, the company suggests, would disregard the Supreme Court's position that copyright exists for the benefit of the general public — not only authors.
On the output side, the company maintains that Claude contains "copyright-protective" features and that it cannot be held responsible when users find workarounds. "When a system 'creates the copy only in response to the user's command,' its operator does not engage in volitional conduct," Anthropic argued in a court filing. "The user does."
The company's position carries the weight of a partial precedent: in the parallel book-publishers case, retired Judge William Alsup ruled that using copyrighted books to train AI was fair use — while simultaneously allowing claims to proceed over Anthropic's use of pirated library copies. That case ended in a roughly $1.5 billion settlement, the largest AI copyright resolution to date.
A Split Bench: Alsup, Chhabria, and the Third Circuit
The legal landscape around AI training remains fractured. Judge Alsup's ruling supported Anthropic on training; two days later, Judge Vince Chhabria ruled for Meta in a similar dispute but cautioned that training may still not constitute fair use "in many circumstances" — leaving the doctrine's boundaries unsettled.
Adding pressure from the other direction, the Court of Appeals for the Third Circuit ruled on September 30 against the fair use defense in Thomson Reuters v. ROSS Intelligence, a case involving a legal research AI trained on Westlaw headnotes. The publishers cited that decision in a brief submitted at the end of September, hoping it galvanizes their position.
Judge Lee is familiar with the dispute: in March 2025, she refused to block Anthropic from using lyrics for training, saying the publishers were asking her to define a licensing market while fair use remained unsettled. Her eventual ruling in the current phase could force that question.
What's at Stake for the AI Industry
The outcome is expected to influence dozens of lawsuits by authors, news outlets, and visual artists against Anthropic, OpenAI, Meta, and other technology companies. A ruling that lyric training is fair use would strengthen the industry's core legal defense; a ruling against Anthropic would accelerate the shift toward licensed training data — and potentially expose AI companies to damages measured in the billions.
For the music industry, the case is also about a licensing market that barely exists today. Lyrics websites pay fees to republish songs; publishers argue that free AI output makes those licensed channels obsolete. For the publishers, the economics are concrete. Lyrics websites that legally republish song texts pay licensing fees to do so, and Concord argues that getting the same words for free through an AI assistant renders those licensed channels obsolete — revenue that flows away from the rights holders who own the underlying works. Whether that market-dilution argument persuades Judge Lee may determine whether AI companies must finally pay for the creative works that underpin their models.
Source note: This article draws on reporting from The News International, The Epoch Times, and Reuters coverage of the October 21, 2026 hearing before Judge Eumi Lee in the US District Court for the Northern District of California.---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
