Udio asks a New York court to seal the number of audio files it trained on

Udio wants one number kept out of public view: how many audio files it used to train its AI music model. On Monday, June 1, the company filed a motion in the US District Court for the Southern District of New York to seal that figure inside the copyright case brought by Sony Music.

Udio, owned by Uncharted Labs, told Judge Alvin K. Hellerstein it wants to redact a single category of information from Sony’s recent filings: what its lawyers call the “Training Data Number.”

The one figure Udio wants kept quiet

The request is narrow by design. The memorandum describes the Training Data Number as the total volume of audio files Sony claims Udio used to train its generative models. In a footnote, Udio added that it does not concede the figure Sony alleges is even accurate.

What Udio is not sealing matters as much as what it is. The company told the court it is not trying to hide the identity of any specific recording, nor the additional works Sony identified by inspecting Udio’s training data.

“

Udio does not seek to seal any particular recording Plaintiffs allege was included in the training data. Rather, the sole information Udio seeks to impound is the total number of audio files allegedly used to train its model.

Udio’s argument is competitive. It told the court the volume of its training data is a core component of the tool it built, in a field crowded with companies racing to ship rival products. If competitors learned that number, the filing says, they could build their own systems faster and more cheaply.

Udio follows Suno’s lead

The move lands days after Suno made the same request in its own copyright case before a federal court in Boston. Suno filed on May 29 and used near-identical reasoning: that rivals could replicate and benchmark against its model.

Two competitors who agree on little have now asked two different courts, in the same week, to hide the exact same figure for the exact same reason. Inner City Press and others have pushed back in the Suno case, arguing the number is a matter of public concern.

Why the count carries weight

The size of a training set is not a footnote in an AI copyright case. It is the scale of the alleged copying, stated as one clean number.

Sony has already moved to assert 30,304 specific recordings it identified through audio fingerprinting, calling that figure only a fraction of the matches it found. Universal and Warner both settled with Udio in late 2025 and turned the fight into licensing deals. Sony is the holdout, and a loss on fair use could push Udio fully onto a licensed model.