On October 18, 2023, three major music publishers (Universal Music Group, Concord Music Group and ABKCO, collectively referred to as the Music Publishers) filed a Complaint and Demand for Jury Trial against Anthropic in the United States District Court for the Middle District of Tennessee Nashville Division. The Music Publishers launched the action to address “systematic and widespread infringement of their copyrighted song lyrics” by the AI company, Anthropic.
In the Complaint, the Music Publishers explained that, in the process of building and operating AI models, Anthropic unlawfully copied and disseminated vast amounts of copyrighted works, namely the lyrics to myriad musical compositions owned or controlled by the Music Publishers. In fact, the Music Publishers argued that Anthropic was supposed to abide by well-established copyright laws, just as countless other technology companies regularly did. They highlighted that it was clear that defendants were not allowed to reproduce, distribute, and display someone else’s copyrighted works to build their own businesses unless they got permission from the rightsholder (this foundational rule of copyright law dated back to the Statute of Anne in 1710).
They say that the United States Copyright Act (17 USC) was violated by Anthropic. The Music Publishers acknowledged that AI was a new technology, but they insisted that AI companies still had to follow the law. Technological advances could not come at the expense of the creators who essentially served as the backbone for AI’s development.
This was how it worked: Music Publishers owned or controlled the copyrights to the musical compositions, including the lyrics. The Music Publishers represented and advocated for thousands of songwriters—they fostered the creation of musical compositions by composers and lyricists, promoted their works, protected their copyrights, and ensured that they received proper remuneration through commercial licensing of the copyrighted works.
Anthropic was using Claude 2 (its newest version is Claude 2.1) for developing, operating, selling, and licensing AI technologies. Anthropic built AI models by scraping and ingesting massive amounts of text from the internet (and potentially other sources), and using all of it to train its AI models and generate output based on this copied text. The Music Publishers claimed that Anthropic copied the data to fuel its AI models lyrics to their musical compositions. They urged that copyrighted material was not free just because it could be found on the internet—in this case, Anthropic never asked for permission (the licensing process was never used).
Even though Anthropic had a company Constitution, where the goal was to act in a way that was harmless, respectful, and ethical, the Music Publishers passionately argued that Anthropic committed copyright infringement since it generated identical or nearly identical copies of their lyrics. In the Claim, the Music Publishers provided examples where famous songs were either completely or partially used in the response to user prompts.
Moreover, the Music Publishers pointed out that there were some websites that acted as lyric aggregators—but they properly licensed the Music Publishers’ copyrighted works and the creators were duly compensated.
In the Claim, the Music Publishers also explained that Anthropic’s AI models generated output containing Music Publishers’ lyrics, even when the models were not specifically asked to do so:
“…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”
In fact, the Music Publishers argued that:
- Anthropic directly infringed the Music Publishers’ exclusive rights as copyright holders, including the rights of reproduction, preparation of derivative works, distribution, and public display.
- Anthropic unlawfully enabled, encouraged, and profited from massive copyright infringement by its users, so it was secondarily liable for the infringing acts of its users under well-established theories of contributory infringement and vicarious infringement.
- Anthropic’s AI output often omitted critical copyright management information regarding these works, making it so that the composers of the song lyrics frequently did not get recognition for being the creators of the works that were being distributed.
The Music Publishers stated, “It is unfathomable for Anthropic to treat itself as exempt from the ethical and legal rules it purports to embrace”. According to the Music Publishers, there was no doubt that Anthropic profited from the infringement of the Music Publishers’ repertoires, since Anthropic was already valued at $5 billion, received billions of dollars in funding, and boasted about numerous high-profile commercial customers and partnerships. The Music Publishers stated in the Claim:
“None of that would be possible without the vast troves of copyrighted material that Anthropic scrapes from the internet and exploits as the input and output for its AI models”
The Music Publishers noted that nothing about Anthropic was creative, but Anthropic depended on the creativity of others and paid them nothing. This caused substantial and irreparable harm.
The Claim set out how Anthropic trained the data:
- copied massive amounts of text from the internet (and potentially other sources), by “scraping” (or copying or downloading) the text directly from websites and other digital sources onto Anthropic’s servers, using automated tools, such as bots and web crawlers, and/or by working from collections prepared by third parties.
- cleaned the copied text to remove material that it perceived as inconsistent with its business model, whether technical or subjective in nature (such as de-duplication or removal of offensive language).
- copied the massive “corpus” and processed it in multiple ways and trained the Claude AI models (encoding the text into tokens).
- processed the data further to fine-tune the Claude AI model and engaged in additional reinforcement learning, based both on human feedback and AI feedback, all of which may require additional copying of the collected text.
The following are Claims for Relief:
- Count I: Direct Copyright Infringement
- Count II: Contributory Infringement
- Count III: Vicarious Infringement
- Count IV: Removal or Alteration of Copyright Management Information
To that end, the Music Publishers requested relief against Anthropic in the form of Judgement on each of the claims above, an order for equitable relief, an order requiring Anthropic to pay the Music Publishers statutory damages, an order requiring Anthropic to provide an accounting of the training data and methods (and the lyrics on which it trained AI models), an order to destroy (under Court supervision) all infringing copies of the Music Publishers’ copyrighted works, costs, and interest.
Is this the end of the story? No, it is not. On November 17, 2023, the Music Publishers asked the Federal Court to stop Anthropic from using their music lyrics; in other words, they asked for a preliminary injunction.
In particular, the Music Publishers argued the following:
- They were likely to succeed on their direct copyright infringement claim (the Music Publishers owned or controlled valid copyrights in the works, Anthropic copied the works, Anthropic’s infringement did not constitute Fair Use—see section 107 of the Act here).
- They would suffer irreparable harm without the injunction (Anthropic deprived the Music Publishers and songwriters of control, denied them credit and goodwill, harmed their reputations, eroded the value of the licensing market, damaged their position to negotiate future licenses, and harmed the relationships that they established between themselves).
- The balance of equities favoured an injunction.
- Enjoining Anthropic’s infringement would serve the public interest.
- The court should not require the Music Publishers to post security
This Claim and recent motion for injunctive relief has come in the midst of other similar disputes, with AI companies, including the Authors Guild commencing and action against OpenAI.
When it comes to music copyrights, time will tell what kind of relief the Music Publishers receive.
It is important to note that this is an American case. In Canada, we have the Copyright Act. We talk about Fair Dealing, not Fair Use.
We will keep you posted on further developments.
Additionally, find Christina at https://www.christinacatenacci.com/
Furthermore, Christina is a co-founder of a new AI company, voyAIge strategy. Check it out at:
https://www.voyaigestrategy.com/
- A closer look at the decision: Google abused its monopoly power - August 28, 2024
- How does ISO/IEC 42001 impact AI governance? - July 19, 2024
- Privacy Commissioner launches investigation into 23andMe data breach - June 20, 2024
Leave a Reply