Universal and Sony are still suing Suno, and have accessed lots of info about the AI company in the process. Suno is very keen for one bit of data, the total number of audio files it used when training its AI, to stay confidential, despite a New York journalist’s efforts to force full transparency
Suno is very keen indeed to ensure that you don’t know how many audio files it used when training its music-generating AI model. Lawyers at Universal Music and Sony Music already know that number, but it nevertheless remains “commercially sensitive information”, according to Suno’s attorneys.
And if you were to ever find out what that number is, for example in court documents filed by the two majors, that would definitely cause “competitive harm for Suno”. And nobody wants that, surely?
Interestingly, Universal and Sony are happy to keep that number secret at Suno’s request. They really want to see a copy of Suno’s licensing deal with Warner Music as part of their legal battle with the AI company, but they are fine with keeping key facts and figures they’ve already accessed as part of these court proceedings confidential, and therefore redacted if included in court filings.
However, New York-based journalist Matthew Russell Lee has formally objected to that plan, reckoning full transparency in this case is in the public interest.
Lee - an investigative journalist who posts articles on his website Inner City Press - last month sent a letter to the court where Universal and Sony are suing Suno for copyright infringement. He urged the judge to block efforts by Suno to keep confidential some of the information it was forced to share with the two music companies during the discovery phase of this dispute.
Setting out his argument for full transparency, Lee says “this is not an ordinary copyright case”, because “the central question - whether and how AI companies may lawfully use copyrighted works to train generative AI models - is among the most consequential legal issues of this era”.
“The answer will shape the future of the music industry, the AI industry, and the rights of artists, composers and copyright holders across every creative field”, he goes on. And “millions of musicians, songwriters and listeners have a direct stake in this litigation”.
All of which is true, though whether that alone is reason to block Suno’s attempts to keep certain facts and figures out of the public domain is debatable. However, Lee reckons that rights set out in the US Constitution regarding public access to court documents “exist precisely for cases of this magnitude”.
He then concludes that “the copyright questions presented in this case will be decided in large part on the factual record now being developed”, and therefore “the public, including working musicians, AI developers and every person who interacts with AI-generated content, deserves to understand the evidence and arguments that will determine those questions”.
Suno's legal team responded to Lee’s letter last week, claiming that the journalist is overstating what information the company is trying to keep out of the public domain. “To be clear, and contrary to Inner City Press’s claims”, they write, Suno is not seeking to hide information about “any particular recording” Universal or Sony believe was used as part of its AI training processes.
And that includes information about the additional 60,000+ allegedly infringed tracks the two majors now want to add to their lawsuit. Both Universal and Sony say that they now know more about what recordings Suno copied without licence when training its AI model and want to list more of them in their lawsuit. Mainly because identifying more works could result in higher damages if they win the copyright case.
The “sole information” Suno wants to keep secret, the AI company’s lawyers go on, “is the total number of audio files allegedly used to train its model”. The lawyers continue, “Suno has never disclosed that figure publicly, and for good reason”.
Generative AI is “a booming field”, they explain, and if any rival AI start-ups “learned the volume of Suno’s training data” they “could use that information to replicate and benchmark against Suno’s models”.
Or they could “infer aspects of Suno’s training and development process” and “potentially optimise their models to unfairly compete with Suno’s by leveraging Suno’s confidential business information”.
Yeah, maybe. Though, given Suno has been quite open about the fact it pulled down huge quantities of music from YouTube, you have to wonder how much more you need to know to train a Suno competitor beyond “steal everything that isn’t nailed down”.
Either way, it doesn’t necessarily feel like that one big number remaining confidential is really all that much of an affront to the public interest. Though the fact Suno is so very keen to keep it secret is probably a good reason why we should get to know it after all.