Music labels clash with Claude over copyrighted lyrics

Anthropic says Claude users ask the chatbot to produce song lyrics and argues that the company is only indirectly responsible for the use of copyrighted works.

SAN JOSE, Calif. (CN) — In an important legal fight over whether artificial intelligence can be trained on copyrighted material, major music publishers warned on Wednesday that Anthropic’s AI chatbot Claude could take away all the music industry’s income from licensing song lyrics.

During a hearing on both sides’ requests for a decision without a trial, Concord Music Group lawyer Matthew Oppenheim said Claude reproduces copyrighted material, sometimes word for word. He called this a violation involving “one of the most highly creative, expressive uses under copyright.”

Oppenheim focused on two kinds of content. The first was “derivative” works, or mashups, where Claude uses copyrighted lyrics to create answers that mention or include parts of those lyrics. The second was “synthetic” works, where Claude creates a new song in the same genre or style as an artist’s work. He argued that both violate the Copyright Act. He also said that the “synthetic” songs Claude creates compete with songs by human artists and weaken the music market.

“It’s not curing cancer, it’s to create other lyrics,” he said. “It couldn’t do it if it didn’t have the lyrics in the database.”

Concord Music Group and other plaintiffs, including Capitol CMG, Universal Music, Songs of Universal, Polygram Publishing and Abkco Music, sued Anthropic in October 2023. They argued that the company’s use of copyrighted lyrics to train its AI was not justified by the growth of the technology industry. They want Anthropic to get a license for copyrighted works if Claude is going to use them.

The music publishers also say Anthropic removes information that identifies the owners and management of copyrighted content to hide its alleged infringement when Claude produces song lyrics word for word.

U.S. District Judge Eumi Lee asked Oppenheim whether his argument would give music publishers control over lyrics so broad that no one, including AI systems, could create new works based on the style of an older copyrighted work without paying for a license to reproduce it.

“Copyright is a monopoly by nature,” Oppenheim said. “That’s not a bad word in the context of copyright.”

Anthropic lawyer Sonal Mehta argued that her client is only indirectly responsible for the use of copyrighted works because Claude’s users are the ones who “push the button” by asking for song lyrics or requesting a song like Taylor Swift’s music. She also said the plaintiffs misunderstand how large language models work. According to Mehta, these models do not take specific lyrics alone but instead learn from “vast, vast amounts” of content.

“It allows it to make probabilistic determinations based on how human language maps,” she said. “Every input weighs on how the model will predict.”

Mehta also argued that the “synthetic” songs Claude creates are new works that can receive copyright protection under the law and do not replace music created by human artists.

“There isn’t anything in the record that shows doom for the music industry,” she said. She said experts had found that Claude caused no harm to the market for music licenses and did not affect licensed lyric search services such as Genius and LyricFind.

In January 2025, Lee ordered Anthropic to keep the existing copyright protections for material entered into and produced by Claude and to apply those protections to future models. However, the judge, who was appointed by Joe Biden, did not broadly ban Anthropic from training its AI on copyrighted lyrics. Last October, Lee rejected Anthropic’s request to dismiss the case. She agreed with the music publishers that Anthropic makes money from the disputed use of lyrics. “Anthropic is paid every time … end users submit a request for publishers’ song lyrics, and it is paid again every time its Claude API generates output copying and relying on those lyrics,” Lee wrote in the decision rejecting the request.

This is not the first time Anthropic has faced legal action over its AI. In 2025, Reddit sued the company, claiming Claude had been trained on Reddit users’ data without permission. Anthropic also settled for $1.5 billion in a major case brought by book authors. The authors said the AI system had downloaded and stored their copyrighted works. It was the largest settlement in the history of U.S. copyright cases.

After several hours of arguments, Lee said she would consider the matter before making a decision. She did not indicate how she would rule on either side’s request for a decision without a trial.