Over the weekend, Bloomberg broke the news that the Sony Music, Universal Music Group and Warner Music Group are in talks with Suno and Udio to license their music to the artificial intelligence startups. If the deals go through, they could help settle the major music companies’ massive copyright infringement lawsuit against Suno and Udio, filed last summer.
Billboard confirmed that the deals in discussion would include fees and possible equity stakes in Suno and Udio in exchange for licensing the music — which the two AI firms have already been using without a license since they launched over a year ago.
That sounds like a potentially peaceful resolution to this clash over the value of copyrighted music in the AI age. But between artist buy-in, questions over how payments would work and sensitivities on all sides, the deals could be harder to pull off than they seem. Here’s why.
You need everyone on board
Ask anyone who’s tried to license music before: it’s a tedious process. This is especially true when a song has multiple songwriters, all signed to different companies — which is to say, almost all of pop music today. Since any music that is used as training data for an AI model will employ both its master recording copyright and its underlying musical work copyright, Suno and Udio cannot stop at just licensing the majors’ shares of the music. They will also need agreements from independent labels and publishers, too, to use a comprehensive catalog.
And what about the artists and songwriters signed to these companies? Generative AI music is still controversial today, and it is foreseeable that a large number of creatives will not take too kindly to their labels and publishers licensing their works for AI training without their permission. One can imagine that the music companies, to avoid a revolt from signees, would allow talent to either opt-out of or opt-in to this license — but as soon as they do that, they will be left with a patchwork catalog to license to Suno and Udio. Even if a song has one recording artist and five songwriters attached to it, it only takes one of those people to say no to this deal to eliminate the track from the training pool.
Is the expiration date really the expiration date?
Licensing music to train AI models typically takes the form of a blanket license, granted by music companies, that lasts between one and three years, according to Alex Bestall, CEO of Rightsify, a production music library and AI company. Other times it will be done in perpetuity. Ed Newton-Rex, former vp of audio for Stability AI and founder of non-profit Fairly Trained, previously warned Billboard that companies that license on a temporal basis should look out for what happens when a deal term ends: “There’s no current way to just untrain a model, but you can add clauses to control what happens after the license is over,” he said.
Attribution technology seems great — but is still very new
Many experts feel that the best way to remunerate music companies and their artists and songwriters is to base any payouts on how often their work is used in producing the outputs of the AI model. This is known as “attribution” — and while there are companies, like Sureel AI and Musical AI, out there that specialize in this area, it’s still incredibly new. Multiple music industry sources tell Billboard they are not sure the current attribution models are quite ready yet, meaning any payment model based on that system may not be viable, at least in the near term.
Flat-fee licenses are most common, but leave a lot to be desired
Today, Bestall says that flat-fee blanket licenses are the most common form of AI licensing. Given the complexities of fractional licensing (i.e., needing all writers to agree) with mainstream music, the AI music companies that are currently licensing their training data are typically going to production libraries, since those tend to own or control their music 100%. It’s hard to know if this model will hold up with fractional licensing at the mainstream music companies — and how they’ll choose to divide up these fees to their artists.
Plus, Mike Pelczynski, founder of music tech advisory firm Forms and Shapes and former head of strategy for SoundCloud, wrote in a blog post that “flat-fee deals offer upfront payments but limit long-term remuneration. As AI scales beyond the revenue potential of these agreements, rights holders risk being locked into subpar compensation. Unlike past models, such as Facebook’s multi-year deals, AI platforms will evolve in months, not years, leaving IP holders behind. Flat fees, no matter how high, can’t match the exponential growth potential of generative AI.”
There’s still bad blood
The major music companies will likely have a hard time burying the hatchet with Suno and Udio, given how publicly the two companies have challenged them. Today, Suno and Udio are using major label music without any licenses, and that defiance must sting. Suno has also spoken out against the majors, saying in a court filing that “what the major record labels really don’t want is competition. Where Suno sees musicians, teachers and everyday people using a new tool to create original music, the labels see a threat to their market share.”
Given that context, there is a real reputational risk here for the labels, who also represent many stakeholders with many different opinions on the topic — not all of them positive. For this licensing maneuver to work, the majors need to be able to feel (or at least position themselves to look like) they came out on top in any negotiation, particularly to their artists and songwriters, and show that the deals are in everyone’s best interests. It’s a lot to pull off.
