“This copyrighted material is not free for the taking simply because it can be found on the internet.”
In the latest example of the entertainment industry pushing back against AI, Universal Music Group — one of the largest record labels in the world — as well as Concord Music Group and ABKCO have sued the Amazon-backed AI startup Anthropic.
The lawsuit, filed in the Tennessee federal court on Wednesday, accuses Anthropic of illegally copying and distributing copyrighted lyrics that were used to train its chatbot Claude 2.
When directly prompted, Claude 2 can spit out the lyrics to hundreds of songs in the publishers’ catalog, ranging from artists like The Rolling Stones, to Beyoncé, to Katy Perry.
“Anthropic builds its AI models by scraping and ingesting massive amounts of text from the internet and potentially other sources,” the publishers wrote in the lawsuit, claiming that included “innumerable” amounts of their musical compositions.
“This copyrighted material is not free for the taking simply because it can be found on the internet,” they added.
A Poor Poet
Even when the Antrophic’s chatbot isn’t directly asked for an existing song’s lyrics, the music publishers allege, Claude 2 will reproduce nearly identical lyrics when prompted to write a supposedly original poem or musical composition in a certain style.
An example provided in the lawsuit: when Claude is asked to “write a short piece of fiction in the style of Louis Armstrong,” the chatbot’s output is almost word-for-word the lyrics to “What a Wonderful World.” (More blatantly, Claude also ripped the title of its “piece of fiction” directly from Armstrong’s iconic song.)
The record labels draw a line between websites that share music lyrics and Anthropic’s AI. Those lyrics aggregators, they point out, often receive licenses and are required to properly give credit.
“But when Anthropic’s AI models regurgitate publishers’ lyrics, they are often unaccompanied by the corresponding song title, songwriter, or other critical copyright management information,” they wrote.
It seems Claude has a guilty conscience, too. According to the publishers, the chatbot will sometimes refuse to output anything using real song lyrics, telling users that that would violate “copyright restrictions.” This demonstrates, the lawsuit alleges, that Anthropic is capable of programming guardrails into the AI.
Among its peers, UMG has been notoriously hostile to AI-generated music.
In April, it demanded that streamers take down a song that used AI-cloned vocals of the rapper Aubrey “Drake” Graham and singer-songwriter Abel Makkonen “the Weeknd” Tesfaye.
And yet, UMG has also dabbled in AI ventures of its own, such as licensing its artists’ melodies and voices for use in Google’s AI Music generator. So the label seems to be fine with AI — so long as it gets paid.
“Universal uses AI in its business and production operation,” the publishers wrote. “Anthropic’s copyright infringement is not innovation; in layman’s terms, it’s theft.”
Is it theft? The decisions of the courts will have to bear that out, in what could be a major precedent in the ongoing copyright battle over generative AI.