Anthropic Fails to Get Lyrics Copyright Infringement Case Dismissed

In a case involving alleged copyright infringement by the Claude generative artificial intelligence (GenAI) tool, a US district court has denied Anthropic's motion to dismiss.

As Music Business Worldwide reports, the decision “keeps alive key portions of a lawsuit filed in 2023 alleging infringement of lyrics from over 500 songs by artists including Beyoncé, the Rolling Stones, and the Beach Boys.”

The lawsuit seems to have been the first over AI’s use of song lyrics and the first against Anthropic.

The Guardian noted that the songs at issue range from the Beach Boys’ God Only Knows and the Rolling Stones’ Gimme Shelter to Mark Ronson and Bruno Mars’ Uptown Funk and Beyoncé’s Halo.

The Plaintiffs (also called “Publishers” in the ruling) are Universal Music Group, Concord Music Group, ABKCO, and five other music publishing companies that own or control exclusive rights to millions of musical compositions.

According to the Guardian,

The publishers also say that Claude illegally reproduces the lyrics by request, and in response to “a whole range of prompts that do not seek Publishers’ lyrics”, including “requests to write a song about a certain topic, provide chord progressions for a given musical composition, or write poetry or short fiction in the style of a certain artist or songwriter”.

For example, the lawsuit said that Claude would provide relevant lyrics from Don McLean’s "American Pie" when asked to write a song about the death of rock pioneer Buddy Holly.

The Plaintiffs assert claims for direct and secondary copyright infringement and for the removal of copyright management information (“CMI”).

According to the plaintiffs, Anthropic used Plaintiffs’ copyrighted song lyrics to train the Claude GenAI model, resulting in the unauthorized dissemination of those lyrics to third-party users.

The Plaintiffs allege that Anthropic used third-party datasets, such as “Common Crawl” and “The Pile,” which contain a significant number of copyrighted lyrics scraped from various websites.

In its response to Anthropic’s motion to dismiss, the court explained that

Contributory infringement liability applies where the defendant “(1) has knowledge of another’s infringement and (2) either (a) materially contributes to or (b) induces that infringement.”

In the context of online platforms,

a computer system operator can be held contributorily liable if it ‘has actual knowledge that specific infringing material is available using its system’ and can ‘take simple measures to prevent further damage’ to copyrighted works, yet continues to provide access to infringing works.

Anthropic contends that the complaint “includes no plausible factual allegation that Anthropic knew of or was willfully blind to any specific infringing lyrics made available to Claude users, or of any third party who had prompted Claude to output infringing lyrics.”

The Plaintiffs raised five arguments to show that Anthropic knew, or had reason to know, of the specific acts of infringement committed by Claude.

For example, Plaintiffs argued that the Court could infer that Anthropic had “actual or constructive knowledge” that “Claude users infringed specific lyrics” based on Anthropic’s implementation and development of “guardrails”:

Publishers assert that Anthropic can program “guardrails” that regulate Claude’s output of copyrighted lyrics, including Publishers’ lyrics….. Publishers claim that, in developing and improving these guardrails, Anthropic “collected Claude user prompts and output data, including specific infringing output copying lyrics.” And Anthropic supposedly became aware of the user prompts seeking such lyrics because each time this occurred, the guardrails detected the users’ actions, and Anthropic received a corresponding notice... Publishers also contend that in conducting a broader study of Claude users’ behavior, Anthropic analyzed “specific efforts by users to avoid these guardrails and identified instances in which the guardrails have failed and Claude has generated infringing output, including output copying Publishers’ lyrics.”

For the sake of the motion to dismiss, the court accepted the Plaintiff’s allegations as accurate.

Reuters reported that in January, Anthropic and the Plaintiffs reached an agreement to maintain existing guardrails and apply them to future Claude models.

As for vicarious copyright infringement, the court noted that

To state a claim for vicarious copyright infringement, a plaintiff must allege that the defendant has (1) the right and ability to supervise the infringing conduct and (2) a direct financial interest in the infringing activity.

In other words, said the court, a defendant “infringes vicariously by profiting from direct infringement while declining to exercise a right to stop or limit it.”

Anthropic argued that Publishers fail to allege that Anthropic has a direct financial interest in Claude users’ alleged infringement.

The Publishers alleged that 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.”

Also, the court found plausible that the availability of Publishers’ lyrics draws customers to use Claude because, as Publishers contend, Claude would not be as popular and valuable as it is but for “the substantial underlying text corpus that includes Publishers’ copyrighted lyrics.”

As for the CMI claim, the Digital Millennium Copyright Act (“DMCA”) restricts the removal or alteration of [CMI] – information such as the title, the author, the copyright owner, the terms and conditions for use of the work, and other identifying information outlined in a copyright notice or conveyed in connection with the work.”

The court inferred that Anthropic had a reasonable basis to know that its intentional removal of CMI from the datasets used to train Claude would conceal its alleged infringement.

Said the court:

Anthropic’s cofounders elected to use a tool, Newspaper, that effectively removed CMI from these datasets—and they chose it specifically because it was more effective at removing CMI than another tool, jusText.

This decision comes a few weeks after another US District Judge approved a $1.5 billion settlement to end a class action brought by authors, journalists, and publishers against Anthropic. That case involved using pirated copies of copyrighted material to train Claude.

According to Music Business Worldwide, the Anthropic settlement could be helpful for the Plaintiffs in the music case:

Crucially for the music publishers, lawyers discovered that LibGen “contains well over a thousand illegal copies of sheet music, songbooks, and other lyric-related books,” including works specifically involved in their lawsuit such as Tiny Dancer (written by Elton John and Bernie Taupin), A Thousand Miles (written by Vanessa Carlton), and 7 Rings (recorded by Ariana Grande).

The music publishers alleged earlier this month that Anthropic hid that it used BitTorrent to pirate these materials, only discovering this through a separate authors’ lawsuit.

“Inexplicably, Anthropic never disclosed to publishers in this case that it had used BitTorrent to copy books containing their works from pirate sites in this manner, despite publishers’ discovery requests calling for exactly this type of information,” lawyers for the music publishers wrote in their recent court filing.

The music publishers are now seeking to amend their complaint to include new charges against Anthropic for distributing copyrighted lyrics without a license, not just using them for training.

Categories: Copyright, Technology