Elvis Presley’s Rights Holder Sues Anthropic Over 274 Songs Used in Training

October 26 — The holder of Elvis Presley’s music catalogue is suing the AI company Anthropic, accusing it of using Presley’s songs to train AI models without permission.

Authentic Brands Group, the licensing company that holds the rights to Presley’s music and likeness, alleges that Anthropic infringed rights in 274 Presley songs. The suit was filed in court last Friday local time.

AI companies have already faced copyright suits from the US Recording Industry Association of America and from labels including Sony, Universal and Warner, but this is one of the first major legal actions brought directly by an artist’s estate against an AI developer. The outcome could set a general precedent for how AI companies may use existing music catalogues.

Lead counsel Richard Busch wrote in the complaint: “The plaintiffs bring this action to protect their proprietary rights. The defendant’s entire business model is built on copyright infringement on an unprecedented scale.”

The complaint states: “From its inception, Anthropic’s AI model, Claude, has been able to generate infringing content from virtually any song in catalogues across the internet.”

The complaint further alleges that Anthropic used 274 Presley songs to train its AI models, including “Can’t Help Falling in Love,” “All Shook Up” and “A Little Less Conversation.” The plaintiffs are seeking $150,000 per infringed work (about 1.003 million yuan at current exchange rates), for a total of about $41 million (roughly 274 million yuan).

The complaint says: “Only after being caught did Anthropic set up so-called guardrails to limit the output of infringing musical content, and those guardrails are ineffective; at the same time the company knew the measures could not stop the model from continuing to produce infringing music. That is enough to show that Anthropic knowingly and deliberately continued to infringe the copyright in others’ musical works.”

AI music platforms work by analysing huge numbers of recordings, learning the timbre, structure and style of songs, and then letting users generate AI music through text or voice commands.

Many musicians have repeatedly voiced fears that their music and voices will be taken up by AI models without permission, and that shoddily made AI-generated music will replace human-written songs.

Some AI bots are already imitating real, living singers on music streaming platforms. Last week, the streaming giant Spotify said that over the past year it had removed roughly 50 million such low-quality AI tracks from its recommendations.