UMG, Sony Music oppose Suno’s bid to keep AI training data confidential

The two major record labels, alongside other plaintiffs, argued in recent court filings that Suno should not be allowed to redact figures relating to the scale of its training datasets. The case centres on allegations that Suno used copyrighted recordings without permission to train its AI music generation systems.

Suno, along with rival AI firm Udio, has sought to prevent disclosure of the number of recordings identified during discovery as having been used in training. The companies argue that revealing the figure could provide competitors with insight into their systems and processes.

However, Universal Music Group and Sony Music Entertainment rejected that position, noting that Suno has already publicly stated that its models were trained on “tens of millions” of works. In their filing, the labels argued that providing a more precise figure would not materially harm Suno’s competitive position.

The plaintiffs are also seeking to keep two specific figures unredacted: an internal count derived from audio-fingerprinting analysis and the overall number of recordings allegedly used in training. The fingerprinting figure, they argue, was produced using Audible Magic technology to identify recordings allegedly ingested by Suno’s systems.

According to the filing, the fingerprinting-derived number is necessary to contextualise the scale of the alleged infringement. The labels said there is “no legitimate basis” for concealing it from public view.

They also argued that the broader training figure is relevant to Suno’s legal defence, particularly its claim that its use of copyrighted material constitutes fair use. The plaintiffs said the number would help demonstrate the extent of copying and whether it extends beyond their own catalogues.

Universal Music Group and Sony Music Entertainment further claimed that concerns over commercial harm are overstated, describing the potential impact of disclosure as “at most, marginal”. They also pointed to Suno’s own acknowledgement of large-scale data use in its training process.

“Having publicly acknowledged copying of that magnitude, Suno cannot credibly claim that disclosing a slightly more precise count of the audio files that its AI model trained on would reveal anything competitively sensitive,” the labels said in their filing.

The dispute forms part of broader litigation involving AI-generated music tools and copyright holders, with similar arguments also emerging in related proceedings involving Udio, which is represented by the same legal team as Suno.

A decision from the court on whether the figures will remain confidential is pending.