Sony Music has asked the federal court in Manhattan to refuse Udio's request to redact one figure from the case: the total number of audio files used to train its music model, which Udio's lawyers call the "Training Data Number". Sony's June 8 filing says the number belongs in public because it appears in the proposed amended complaint and bears directly on Udio's fair-use defence.
Udio moved to seal the figure on June 1, arguing that rivals could use it to build competing products faster and cheaper and to infer its training strategy. Chief executive Andrew Sanchez said in a declaration that dataset size is closely guarded because it reveals choices about model design. Sony called those claims speculative and said a single aggregate number reveals nothing about how Udio built its models, only the scale of what it copied.
The fight is separate from Sony's bid to add about 30,000 recordings found in Udio's training data during discovery; Udio is not trying to seal that list. Judge Alvin Hellerstein had earlier granted Udio's sealing request, then vacated that order on June 3 and asked for full briefing. A near-identical dispute is playing out in the Massachusetts case, where Universal and Sony oppose Suno's attempt to seal its own training-data size.
Sony is the only major still suing Udio after Universal and Warner settled in late 2025. Udio has admitted taking audio from YouTube for training but calls its use fair, and both sides are heading for summary-judgment motions on that question.
Sources · 1