Music publishers say Anthropic's AI copied song lyrics without paying

Major music publishers are asking a federal judge to rule that Anthropic, the maker of the Claude AI assistant, broke copyright law by using song lyrics to train its AI without permission or payment. The case is being heard in San Jose, California, and it could help set the rules for how AI companies use creative work.

What the publishers claim

Universal Music Group, Concord and ABKCO say Anthropic copied lyrics from songs tied to artists including Beyoncé, The Rolling Stones and The Beach Boys. They say the company built a commercial product from that material and skipped the licensing fees that lyric services normally pay.

They also argue that Claude can reproduce or generate text similar to copyrighted lyrics. In their view, that could compete with the market where lyrics are licensed and weaken the reasons to write and publish songs.

Anthropic's defense

Anthropic says training an AI is not the same as republishing a lyric. It says the lyrics are one small part of a huge pile of text used to teach the model patterns in language. Claude, it argues, is not a database of songs.

The company is relying on fair use. That is a part of US copyright law that can allow use of protected material without permission in some cases. Courts look at the purpose of the use, how much was copied and whether it hurts the market for the original.

Why courts are not in agreement

Judge Eumi Lee is deciding whether the training counts as fair use. Other judges have gone different ways. In an earlier Anthropic case with authors, Judge William Alsup said training could be highly transformative, but he found problems with Anthropic's storage of pirated books. That case ended in a $1.5 billion settlement.

In a separate case involving Meta, Judge Vince Chhabria warned that AI could flood creative markets with competing content and reduce the value of human work.

What happens next

Judge Lee's ruling may not settle the issue nationwide, but it could become a major precedent. OpenAI, Meta, Google and several AI startups face similar lawsuits. If fair use is accepted broadly, training can go on without licensing each work. If not, AI companies could face higher costs and more pressure to sign licensing deals, as some already have. Those costs could eventually show up in the AI tools people use.