CLIENT ALERT: For AI Companies, How You Get Your Training Data Matters—SDNY Allows DMCA Claim to Proceed Against AI Music Generator Udio

KEY TAKEAWAYS:

• On April 15, 2026, Judge Alvin K. Hellerstein of the Southern District of New York denied a motion to dismiss a Digital Millennium Copyright Act (“DMCA”) anti-circumvention claim brought by major record labels against Uncharted Labs, Inc., the company behind the AI music generator Udio.
• The DMCA, among other things, prohibits the circumvention of technological measures that control access to copyrighted works. It was designed to protect copyright holders in the digital age by making it unlawful to bypass security technologies that guard copyrighted content.
• Judge Hellerstein held that the plaintiffs stated a viable claim under the DMCA, 17 U.S.C. § 1201(a), based on their allegations that Udio circumvented YouTube’s platform protections to download copyrighted content without authorization and used that content to train its AI model.
• The ruling arrives at a pivotal moment in AI copyright law, as courts grapple with both whether training AI models on copyrighted materials constitutes fair use and whether the methods by which AI companies acquire training data may independently violate the law.

 

BACKGROUND:

Uncharted Labs, Inc., d/b/a Udio (“Udio”) is a generative AI-powered music generation service. Users can enter prompts, including descriptions, genres, lyrics, and themes, to generate music files.

In June 2024, a group of record companies and recorded music businesses filed suit against Udio, asserting claims for copyright infringement. The DMCA anti-circumvention claim at issue was added by amendment in late 2025. Since the lawsuit was filed, Universal Music Group and Warner Music Group have each settled with Udio and announced licensing partnerships, leaving Sony Music Entertainment and its affiliates as the remaining plaintiffs.

The plaintiffs allege Udio trains its AI models on copyrighted sound recordings obtained without permission from copyright holders. Specifically, the plaintiffs allege YouTube employs certain technological measures, including a “rolling cipher,” to control access to content and prevent protected media files from being downloaded, and that Udio acquires copyrighted recordings by circumventing these technological protections to download content without authorization, a practice known as “stream ripping.”

In October 2025, Udio moved to dismiss the DMCA claim, arguing that YouTube’s rolling cipher is not a technological measure that effectively controls access to copyrighted works under the DMCA (i.e., that it functions as a copy control, not an access control, and therefore falls outside the statute’s anti-circumvention provisions). It did not challenge the underlying copyright infringement claim.

THE COURT’S RULING:

Judge Hellerstein denied the motion to dismiss.

The Court explained the DMCA’s framework, including that it prohibits the circumvention of “access controls,” meaning it prohibits the circumvention of technologies designed to prevent access to a work. Critically, Judge Hellerstein explained that the statute does not prohibit the circumvention of “copy controls” (technologies designed to permit access to a work but that prevent copying or other infringement of the work). However, the statute does prohibit trafficking in tools designed to circumvent either access controls or copy controls.

The Court’s decision turned on whether YouTube’s rolling cipher was an access control or a copy control. The Court concluded that the plaintiffs’ allegations that the rolling cipher regulates access to YouTube’s content and Udio circumvented the rolling cipher to download content in bulk sufficed to state a claim for circumvention of technological measures.

However, the Court cautioned that, while the plaintiffs’ allegations were sufficient at the pleading stage, “[w]hether YouTube’s measures ultimately constitute access controls within the meaning of § 1201 requires a greater factual record,” and the Court authorized Udio to renew its arguments later in the case.

The Court also noted the pending appeal in Yout LLC v. Recording Industry Association of America, Inc., No. 22-2760 (2d Cir.), which presents relevant questions regarding the same statutory provisions, but declined to predict the outcome of that appeal.

PRACTICAL IMPLICATIONS:

This ruling arrives at a pivotal moment in AI copyright law. In June 2025, two federal courts in the Northern District of California found that using copyrighted works to train AI models constituted fair use. In Bartz v. Anthropic PBC, No. 24-cv-05417 (N.D. Cal. June 23, 2025), the Court held that Anthropic’s use of copyrighted books to train its Claude AI model was “quintessentially transformative” and therefore fair use (though it simultaneously held that downloading pirated copies of those books was not protected). In Kadrey v. Meta Platforms, Inc., No. 23-cv-03417 (N.D. Cal. June 25, 2025), the Court similarly found that AI training was “highly transformative” and fair use, while cautioning that the ruling was narrow and fact-specific.

Against that backdrop, the DMCA anti-circumvention claim in this case takes on heightened significance: even if AI training on copyrighted material is ultimately deemed fair use, the method by which training data is acquired may independently violate the DMCA. Judge Hellerstein’s ruling suggests that rights holders may be able to seek legal recourse based on how AI companies obtain their training data, separate and apart from how that data is used.

The access control versus copy control distinction at the heart of this ruling is also before the Second Circuit in Yout LLC v. Recording Industry Association of America, Inc., No. 22-2760, which could provide binding guidance on the same statutory question Judge Hellerstein left open for further factual development.

This decision adds to a growing body of caselaw grappling with the rapid expansion of AI usage and model training across various industries. Clients that rely on tools to acquire model training data from streaming platforms or other protected sources should carefully evaluate their practices and potential exposure under the DMCA.

CONCLUSION:

The Court’s ruling underscores that certain methods of procuring data to train an AI model may fall within the DMCA’s purview, opening the door for rights holders to pursue legal recourse. Sher Tremonte LLP will continue to monitor developments. For questions, please contact us. Written by Amanda Greenfield and Emily McDaniels.