Photo Credit: Jason Leung
Suno isn’t alone in pushing to keep its “training data number” – or a tally of the tracks used to train its models – out of the public view. Now, amid a legal battle with Sony Music, AI music rival Udio is also attempting to seal the figure.
Still fending off copyright litigation from Universal Music and Sony Music, Suno was first to request that the evidently important total remain under wraps. According to the platform, the corresponding motion pertains specifically to “the number of audio files” (not the titles thereof) allegedly “used to train its generative AI model.”
And per Suno, competitors could potentially capitalize on the info “to replicate and benchmark against Suno’s model” and “optimize their models to unfairly compete with Suno’s.”
Does this position tell the full story? Or are the companies in actuality looking to prevent the numbers’ release due to optics and legal concerns? Both figures came to light in marathon discovery processes during which the major label plaintiffs had access to the relevant datasets.
Of course, we don’t know precisely what the majors unearthed. But we do know that they – meaning Universal Music against Suno and Sony Music against Suno and Udio alike– promptly supersized their actions.
Consider the point alongside the platforms’ technical capabilities, existing details about different AI giants’ all-encompassing training, the sweeping music infringement allegations in other suits, and related years-old comments from Suno, and it seems safe to describe the total as sizable.
Furthermore, there doesn’t appear to be a huge risk of other AI developers’ using the information “to replicate” the cash-flush Suno’s model – unless they’re unbothered by the distinct possibility of spurring massive infringement suits with free-for-all training.
In any event, regardless of the exact reasons for seeking to seal, both Suno and Udio are pulling out all the stops here.
As many are already aware, each of the companies has tapped Latham & Watkins for representation; as the same attorneys penned the documents at hand, there’s a great deal of overlap between them.
Even so, only Udio’s filing explores the possibility of competitors’ leveraging the tally to develop AI products “more rapidly and at a lower cost”; Suno’s submission contains no mention of cost whatsoever.
“If competitors learned the volume of Udio’s training data—a core component of the tool has [sic] Udio built—they could use that information to develop and refine competing products more rapidly and at a lower cost by using the Training Data Number to measure their own training datasets against Udio’s and extrapolate commercially sensitive insights about Udio’s training methodology and strategy, thereby obtaining an unfair competitive advantage,” the roughly 70-word sentence reads.
Given these and similarly high-stakes suits against Udio and Suno, it’s easy to overlook the ramping-up Warner Music-Suno partnership. Forthcoming music-generation offerings are understandably the chief focus, and we’ve heard comparatively little about topics including the major’s distribution role.
But some will recall Warner Music-owned ADA’s part in distributing Boomy generations en masse. On Reddit, one self-described Suno user recently claimed to have received several YouTube copyright claims from Warner Music and ADA.
Admittedly, there’s a lot going on in this particular example. But in brief, the Redditor is said to have pumped out a “background” track via Suno; incorporated it into video content without uploading to on-demand services; and then seen third parties allegedly lift the music en route to adding vocals and distributing to DSPs via “artist” profiles.
On YouTube, the Redditor allegedly received the copyright claims from ADA, Warner Music proper, and Smart Music Group; “WMG auto-rejected every single” dispute, per the post.
