US judge reopens debate over Udio AI training data

The development comes amid ongoing copyright infringement litigation involving major record companies and artificial intelligence music platforms Udio and Suno. Both companies have sought to prevent the release of what are known as their “training numbers”, figures representing the total number of recordings used to train their generative AI models.

The companies argue that making the figures public could provide competitors with commercially sensitive information. However, the exact training numbers have remained undisclosed throughout the legal proceedings.

The latest development concerns a lawsuit involving Udio and several music companies led by Sony Music Entertainment. According to court filings, Sony Music recently sought permission to file a redacted version of its amended complaint because it contained information that Udio regarded as confidential and suitable for sealing from the public record.

Earlier this week, Judge Alvin K. Hellerstein initially approved Sony Music’s request. However, on the following day, he issued a separate order overturning a previous decision that had granted Udio’s request to seal references to its training numbers.

In the order, Judge Hellerstein stated: “My Order granting Defendants’ motion to seal, ECF No. 166, is hereby vacated. The parties may brief that motion to seal, ECF No. 163, in the ordinary course.”

The ruling does not automatically make the information public but requires the parties to continue arguing the matter before the court reaches a final decision on whether the material should remain confidential.

The dispute centres on information uncovered during extensive discovery proceedings in the copyright lawsuits. The figures are considered significant because they relate to the scale of copyrighted recordings used to train AI music-generation systems.

Suno has previously acknowledged that its training data includes “tens of millions of public music audio files and corresponding textual metadata”, although it has not disclosed a precise number.

The legal battle forms part of a broader dispute between the music industry and AI developers over the use of copyrighted recordings in training artificial intelligence systems. Record labels argue that AI companies used protected works without permission, while AI developers have sought to defend their practices under the principle of fair use.

In a separate filing, Suno has asked the court to focus on its fair use defence rather than allow further expansion of the lawsuit brought by major record companies.

“After two years of extensive fact discovery,” Suno said in a court filing, “Suno is entitled to a timely consideration of its fair use defense. The Court should reject Plaintiffs’ attempt to expand this case beyond that can be manageably litigated, at least without undue and indeterminate delay.”

Further court submissions are expected to provide greater clarity on whether Udio’s training numbers will ultimately remain confidential or be disclosed as part of the ongoing litigation.