Lyric or Leave It: Anthropic Tries to Strike a Chord with the Music Industry

Do These “Guardrails” Strike the Right Note?
On December 30, 2024, Anthropic agreed to a partial injunction while continuing to fight against the copyright infringement claims. In the now-signed stipulation, Anthropic stipulates that it has already implemented “guardrails” to prevent the reproduction of copyrighted works. While the content of the “guardrails” remains redacted at this time, the stipulation states that Anthropic will:

• Maintain current technical measures to prevent AI models, such as Claude, from generating or reproducing content that infringes on copyrighted song lyrics;
• Apply these safeguards consistently to all future large language models and product offerings developed by Anthropic; and
• Establish a protocol for music publishers to notify Anthropic if they believe the guardrails are insufficient. Anthropic is required to investigate and address such concerns promptly, with potential court intervention if needed.

These measures appear to strike a balance between advancing AI technology and protecting intellectual property, particularly by addressing the plaintiffs’ concerns of preventing models from generating outputs that could infringe on copyrighted song lyrics. The concept of these guardrails and this stipulation is reminiscent of legal issues raised in earlier technology cases, such as Google Books. Authors Guild v. Google, Inc., 804 F.3d 202, 212 (2d Cir. 2015). Indeed, Anthropic defends its use of copyrighted lyrics for AI training by invoking the fair use doctrine, citing Authors Guild v. Google, where the court ruled that copying for purposes like creating a searchable database can be fair use, as it expands public knowledge. Anthropic argues that, similar to Google Books, its use of copyrighted lyrics is transformative—training its AI to recognize language patterns without directly reproducing the lyrics. Anthropic further states that its position is also supported by past rulings that found intermediate copying for new technological purposes to be fair use. For now, how the Courts will analyze this pivotal question of “fair use” in this context will be for another day.

Anthropic emphasizes the guardrails it has implemented to prevent copyright infringement, asserting that it will continue to apply these safeguards across all its current and future AI models. Anthropic also commits to investigating and addressing any concerns raised by the Publishers about the effectiveness of these guardrails, with the Publishers retaining the right to seek court intervention if needed. However, key issues remain unresolved, particularly the Publishers’ demand that Anthropic stop using unauthorized copies of their lyrics to train future AI models. Both parties have reserved their rights on these matters, and the Court will maintain jurisdiction to enforce any disputes related to the guardrails. The Court has not yet ruled on Anthropic’s motion to dismiss parts of the lawsuit, although it is expected that dismissal will be granted with leave to amend.

Taking Aim at the Music Industry?
This case is part of a broader trend of lawsuits filed by content owners in the music industry against AI companies. Recently, we discussed cases in which Universal Music Group Records, Sony Music Entertainment, and other major record labels filed two complaints against generative AI startups Suno, Inc. and Uncharted Labs, Inc., similarly alleging copyright infringement due to their use of copyrighted music to train AI models that generate similar-sounding music. Those cases survived a motion to dismiss and are at the discovery stage, it is expected that at least portions of this case will similarly survive.

These music cases also fall on the heels of copyright cases filed by writers, comedians and journalists about AI models. Resolution of these critical copyright issues lay not only in the specifics of each AI model but also in the mechanisms that can be put in place to protect copyright owners from any potential infringement.

Pillsbury is continuing to closely monitor developments in AI-related litigation across industries in the United States. For more information and practical advice for navigating the ever-changing regulatory landscape, please contact the authors of this alert or your regular Pillsbury contact.