Anthropic’s Claude 2 provides users with lyrics to ‘Roar’ upon request, but it does so without a licensing agreement.
Major record label Universal Music Group and other music publishers have filed a lawsuit against artificial intelligence company Anthropic for distributing copyrighted lyrics using its AI model, Claude 2.
Universal Music is Suing AI Company Anthropic for Distributing Song Lyrics
The complaint, which the music publishers filed in Tennessee, alleges that Claude 2 can distribute lyrics that are almost identical to songs such as Katy Perry’s “Roar,” Gloria Gaynor’s “I Will Survive,” and the Rolling Stones’ “You Can’t Always Get What You Want” when prompted.
They also claim that Claude 2’s results feature phrases that are extremely similar to existing lyrics, even when not explicitly instructed to recreate songs. As an example, the complaint cites a prompt like “Write me a song about the death of Buddy Holly,” which resulted in the large language model producing the lyrics to Don McLean’s “American Pie” verbatim.
Sharing lyrics online isn’t a novel concept. Platforms like Genius thrive because individuals frequently forget song lyrics. Nevertheless, music publishers highlight that numerous lyric-sharing platforms pay for licensing these lyrics. They assert that Anthropic “frequently neglects essential copyright management information.”
The complaint states, “Several music lyrics aggregators and websites already fulfill this role, but these platforms have correctly obtained licenses for the copyrighted works of publishers to offer this service. In fact, there’s an established market where publishers license their copyrighted lyrics, guaranteeing that creators of musical compositions receive compensation and acknowledgement for such uses.”
The plaintiffs claim that Anthropic not only distributes copyrighted material without authorization but also uses it to train its language models.
UMG states that it utilizes AI tools in its business and production processes. However, it alleges that by distributing material without permission, “Anthropic’s copyright infringement isn’t innovation; in simple terms, it’s theft.”
The complaint argues that Anthropic can block the distribution of copyrighted material. It also claims that Claude 2 declines to respond to certain prompts that request specific songs due to copyright infringement. UMG, though, did not specify which songs they are referring to.
Legal and Ethical Challenges in AI and Music Industry Partnerships
These responses clearly indicate that Anthropic is aware that generating content that replicates others’ lyrics goes against copyright law. However, even with this knowledge and the apparent ability to control infringement, Anthropic frequently fails to establish effective and consistent safeguards against the infringement of publishers’ works, according to the plaintiffs.
Copyright infringement has become a highly debated topic in generative AI, and the music industry has been grappling with how to utilize the technology while safeguarding its rights. Several lawsuits have been filed against generative AI platforms like ChatGPT, Stable Diffusion, and Midjourney concerning the use of protected data and the creation of results resembling copyrighted material.
UMG publicly stated its collaboration with companies like Google on AI matters, having partnered with YouTube to influence its approach to generative AI on the platform.
Anthropic emphasizes its commitment to trust and safety. Much of its product and research principles are rooted in a concept known as “constitutional AI,” as described by The Verge’s James Vincent, which involves training AI systems to adhere to a predefined set of rules.
In September, Amazon made a $4 billion investment in Anthropic. Other investors include Google, which contributed $300 million to the company.