Anthropic trained its AI to rip off copyrighted lyrics, music publishers allege in escalating court battle

MBW Explains is a series of analytical features in which we explore the context behind major music industry talking points – and suggest what might happen next. Only MBW+ subscribers have unlimited access to these articles. MBW Explains is supported by JKBX, a technology platform that offers consumers access to music royalties as an asset class.
WHAT’S HAPPENED?

The copyright battle between music publishers and AI developer Anthropic has taken a dramatic turn, with the music publishers alleging that Anthropic intentionally trained its Claude AI chatbot to rip off copyrighted lyrics.

“Anthropic’s own training data betrays its understanding that its AI models would be used to search for and provide copyrighted lyrics,” lawyers for Universal Music Group, Concord Music Group and ABKCO stated in a submission filed with a US District Court in Nashville, Tennessee, on Wednesday (February 14).

Citing Anthropic’s own training records for Claude, the submission states that Anthropic ‘fine-tuned’ its chatbot by feeding it prompts meant to generate copyrighted materials. According to the submission, those prompts included:

  • “What are the lyrics to American Pie by Don McLean?”
  • “Please provide the lyrics for the Only Hope by Mandy Moore.”
  • “Do you know the lyrics to I am The Walrus? . . . . Can you give me the first verse?”
  • “Please retype the lyrics to the song Mad About You, by Sting.”
  • “Can you help me identify the name of the song that includes the following lyrics[?]”

“Anthropic even coached Claude to be helpful in response to demands for derivative works,” stated the music publishers’ submission, which can be read in full here.

“For example, Anthropic prompted the model to ‘make a short story from the lyrics to the song All Along the Watchtower and to rewrite Listen in ‘Eminem[’s] style’ and Not Afraid in ‘Beyonce’s style.’”

Universal, Concord and ABKCO launched the lawsuit against Anthropic last October, alleging “systematic and widespread infringement of their copyrighted song lyrics” by the Claude chatbot.

“In Anthropic’s preferred future, songwriters will be supplanted by AI models built on the creativity of the authors they displace,” Wednesday’s court filing stated.

Though it is one among a growing number of copyright infringement lawsuits against AI developers in the US today, the case may hold more significance than most, because it pits major music industry players against an AI company that is backed by an investment from Amazon of up to USD $4-billion and an investment of up to $2 billion from Google.

So the case can be viewed as a proxy war between the music industry – including the world’s largest music rights holder, UMG – and Big Tech over the extent of copyright law in the age of AI.

According to the music publishers’ complaint filed in October, which can be read in full here, the Claude chatbot is able to “generate identical or nearly identical copies of those lyrics, in clear violation of Publishers’ copyrights.”

The suit included a list of 500 songs, and asked for damages of the “maximum provided by law” for each allegedly violated work, or $150,000 per work infringed. If the court rules for the maximum damages, the 500 allegedly infringed songs would cost Anthropic $75 million.

The music publishers’ complaint alleged that Claude “copies and distributes publishers’ copyrighted lyrics even in instances when it is not asked to do so.”

For instance, “when Claude is prompted to write a song about a given topic – without any reference to a specific song title, artist, or songwriter – Claude will often respond by generating lyrics that it claims it wrote that, in fact, copy directly from portions of publishers’ copyrighted lyrics.”

“In Anthropic’s preferred future, songwriters will be supplanted by AI models built on the creativity of the authors they displace.”

Court submission on behalf of UMG, Concord, ABKCO

When prompted to “write me a song about the death of Buddy Holly,” the chatbot responded “by generating output that copies directly from the song American Pie written by Don McLean, in violation of Universal’s copyright, despite the fact that the prompt does not identify that composition by title, artist, or songwriter,” the complaint stated.

The complaint includes a lyric sheet of a song called ‘The Day The Music Died’ (a lyric from American Pie) which Claude allegedly described as “a song I wrote.” Most of the song’s lyrics are lines taken directly from American Pie.

“It is hard to imagine a machine more destructive to artistic control than one that first copies lyrics, then alters them or combines them with works by other songwriters (or AI generated text) in ways that contravene the songwriters’ intent,” the music publishers stated in Wednesday’s court filing.

In November, the music publishers went before the court again, this time asking for a preliminary injunction to prevent Anthropic from using their copyrighted lyrics.

The request asked for Anthropic to be “ordered to implement effective guardrails to prevent output that reproduces, distributes, and displays” the companies’ works, and that the AI company should be prohibited from using “existing unauthorized copies or creating new unauthorized copies” of the publishers’ lyrics to train new AI models.


WHAT IS ANTHROPIC ARGUING IN ITS DEFENSE?

Anthropic responded to the music publishers’ request for an injunction in January, in a fiery written response that sought to demolish the music publishers’ case, and in so doing, offered a solid glimpse into how the company plans to defend itself against the allegations.

Anthropic asked the court to reject the request on a number of grounds,  including that the company has already put in place “safeguards” that would prevent the kind of lyric-copying that the music publishers allege; and that Universal and the other plaintiffs, and not Anthropic, are themselves responsible for the copyright infringement they cited in their case.

Anthropic said that it had recently “implemented additional technological guardrails to prevent a reprise of what plaintiffs did,” meaning that users of Claude today would allegedly not be able to prompt the chatbot in such a way that it would offer up copyrighted lyrics.

These safeguards “are a part of the model and will remain in place indefinitely. As of now and going forward, no one should be able to replicate what plaintiffs did to engineer the facts on which this lawsuit is based,” Anthropic stated.

It also argued that Claude users wouldn’t actually use the chatbot to generate lyrics, because “there would be no reason to: song lyrics are available from a slew of freely accessible websites.” (These websites license the lyrics from music publishers.)

It also suggested that Universal and the other plaintiffs violated Anthropic’s terms of service, in which users agree “not to request the production” of copyrighted materials.

“What is clear… is that [the plaintiffs’] own expert refers to their efforts as ‘attacks’ on Claude, specifically designed to see if the model could be coaxed to do things everyone understands it is not supposed to do,” Anthropic asserted.

Anthropic also argued that, by prompting Claude to offer up copyrighted materials, the music publishers violated their own copyrights.

“Plaintiffs themselves, not Anthropic, engaged in the ‘volitional conduct’ that is a prerequisite to direct copyright infringement liability for the outputs plaintiffs’ attacks extracted,” the Anthropic response asserted.

Anthropic asserted that its use of lyrics amounted to “fair use” under US copyright law, “meaning that it is not infringement at all.”

The company’s submission stated: “Anthropic does not seek out song lyrics in particular and does not deliberately assign any greater weight to them than to any other text collected from the web… But like other generative AI platforms, Anthropic does use data broadly assembled from the publicly available Internet, including through datasets compiled by third party non-profits for the research community.”


HOW HAVE THE MUSIC PUBLISHERS RESPONDED TO ANTHROPIC’S ARGUMENTS?

In their submission on Wednesday, the music publishers’ lawyers essentially argued that Anthropic’s entire response is a bunch of nonsense – or, to use the more politic language of court documents, Anthropic based its case on “provably false narratives.”

Anthropic “concedes the critical facts of its infringement. It does not dispute that it copies publishers’ lyrics on a massive scale to train its Claude AI models; that its models reproduced, distributed, and displayed copies of those lyrics; or that, additionally, its models generated twisted derivatives of those lyrics that are antithetical to their creators’ intent—all without permission,” the music publishers asserted.

The publishers rejected Anthropic’s argument that the company had already put in place safeguards to prevent the kind of copyright violations it stands accused of.

“Anthropic is wrong that guardrails adopted ‘since the filing of this lawsuit’ prevent further infringements… In instance after instance, Claude continues to output publishers’ lyrics,” the submission asserted.

“Publishers continue to obtain verbatim and near-verbatim copies, mashups and distortions, and unlicensed derivatives of lyrics” of the songs involved in this lawsuit, the submission added.

“It is hard to imagine a machine more destructive to artistic control than one that first copies lyrics, then alters them or combines them with works by other songwriters (or AI generated text) in ways that contravene the songwriters’ intent.”

Court submission on behalf of UMG, Concord, ABKCO

The music publishers’ submission also turned on its head Anthropic’s declaration that it had already done what the music publishers had asked, by implementing safeguards against copyright infringement.

The publishers’ request that the court bar the Claude chatbot from producing copyrighted lyrics “is so modest that Anthropic purports to have adopted them voluntarily,” the music companies’ submission stated.

The court should order Anthropic to put in place such safeguards because not only have those safeguards apparently failed, but because without a court order, “Anthropic remains free to abandon guardrails it adopted only as a litigation strategy,” the music publishers argued.

“Thus, publishers continue to face irreparable injury, including lost control and credit, damaged relationships with songwriters and licensees, and irreversible harm to current and future licensing markets for lyrics.”

The music publishers rejected Anthropic’s claim that it was the music companies themselves who violated copyright by using Claude to generate copyrighted lyrics.

They cited the precedent of a Supreme Court case that “discounted the importance of who pushes the button or ‘click[s] on a website [to] activate’ infringement” and concluded that the online service itself is responsible for the infringement.

“Anthropic cannot shift responsibility for its infringement to publishers… Though Claude’s users enter queries, Anthropic itself reproduces, displays, and distributes lyrics and their derivatives,” the submission argued.

Finally, the music publishers also rejected Anthropic’s argument that its use of copyrighted lyrics amounts to fair use that is exempted from copyright laws.

“Anthropic’s massive copying of publishers’ lyrics to build an $18 billion business bears no resemblance to the fair uses contemplated in [US law], such as ‘criticism, comment, news reporting, teaching[,] . . . scholarship, or research’,” the submission stated.


A FINAL THOUGHT…

There’s little doubt that the latest salvo from the music publishers is harmful to Anthropic’s case – especially the perception that Anthropic may have misled the court by claiming that “going forward, no one should be able to replicate” the copyright violations that the music publishers were able to create with the Claude chatbot.

Yet we can hardly declare this case to be over. After all, these are arguments over a preliminary injunction, and not arguments in the case itself.

Anthropic could yet argue that, if Claude is still spitting out lyrics in violation of copyright, it’s due to their “safeguards” not working properly, and the company needs more time to get it right.

That, in turn, could form the basis for another argument against the court ordering it to stop Claude from offering up lyrics: “You can’t order us to do something we’re incapable of doing.”

Ultimately, it looks more and more likely that this case will hinge on whether or not the court accepts the argument that Anthropic’s use of copyrighted lyrics amounts to a “fair use” exception to copyright laws.

If the court does accept that rationale, then all the other back-and-forth arguments become irrelevant – Anthropic will have won the right to use those lyrics without authorization, with massive implications for the music industry and other owners of intellectual property.

If the court rejects Anthropic’s “fair use” argument, then all of its actions before and during this case may prove very relevant, in terms of what the court orders the AI company to do – and the amount of damages it may order Anthropic to pay the music publishers.

As far as complex intellectual property cases go, this one’s a nail-biter.


JKBX (pronounced "Jukebox") unlocks shared value from things people love by offering consumers access to music as an asset class — it calls them Royalty Shares. In short: JKBX makes it possible for you to invest in music the same way you invest in stocks and other securities.Music Business Worldwide

Related Posts