The Music Labels Sued AI for Stealing Their Music. Then They Licensed It to the Same Companies.

Two years ago the biggest record labels went to war with a pair of small startups. In June 2024, Universal, Sony Music Entertainment and Warner Music Group sued the AI music generators Suno and Udio for copyright infringement, in litigation coordinated by the RIAA. The charge was that the two companies had trained their models by copying copyrighted recordings spanning many artists, genres and time periods, without permission or payment. Read today, the complaints look like a defence of artists — a trade taking on the machines on behalf of the people whose music mattered. That makes everything that happened afterward awkward to explain.

The fact that I could do something once and then it could be reused, and if that's not negotiated in a contract, then I never got to make a choice to receive additional compensation. It all comes down to the loss of money.

— Matt Evans, a session musician in Pasadena, in the Los Angeles Times

Because in the eighteen months that followed, the same labels settled with the same companies and, one by one, began licensing them access to the very same catalogs. In October 2025, Universal settled with Udio and became its partner. Warner settled with both Udio and Suno in November. BMG — which was never even a plaintiff — licensed to Suno in August, and Believe followed this month. When Suno released its v6 models on 9 September, they were built on music licensed from Warner, BMG and Believe. The company the labels once described as having built its product on their stolen recordings now works for them.

Let's be fair about the platforms before judging the dealmakers. A text prompt genuinely becomes a finished song in under a minute, and the licensed v6 models produce usable, distributable music for working creators. The product is real. The argument was never honestly about whether it should exist.

The lawsuit was always a negotiation

None of this is hypocrisy in the simple sense. The labels' real position on AI music was never 'artists first.' It was 'we get paid and we keep control.' The lawsuit was leverage; the licence is the payout. An entertainment attorney who specialises in copyright put the arithmetic plainly in the Los Angeles Times.

The reason why they would settle this is they think they're going to profit more from licensing music to them rather than from the outcome of a lawsuit.

— Jane Davidson, entertainment attorney at Nolan Heimann, in the Los Angeles Times

The deeper reason the labels could make this deal without asking anyone is structural, and it explains who ends up with no seat at the table. A song carries two copyrights. One sits with the songwriters and their publishers, over the composition. The other sits with whoever owns the sound recording — and under the standard major-label contract, that is the label. So when an AI company wants to train on the masters, it does not need to ask the artist. It needs to ask the label. The people who wrote and performed the music do not get a vote; they get whatever their own contracts allow, and only if the owner of the masters decides to share.

The part the lawsuits never mentioned

The reminders of who was left out have been coming from the people the 2024 lawsuits claimed to protect. In June, the American Federation of Musicians sued Universal and Warner, alleging the labels had licensed recordings its members played on to Suno and Udio without paying those musicians or even telling the union which recordings were involved. A solo a session player recorded once, years ago, is now training data — and under many contracts, the decision to sell that use was the label's, not theirs.

By licensing our members' performances to AI platforms like Suno and Udio without consent, credit, or compensation, companies like Universal, Warner Records, and Atlantic are bypassing the very human creators who made their catalogs valuable in the first place.

— Tino Gagliardi, president of the American Federation of Musicians, in a statement

The songwriters have their own complaint, and it is about the half of the copyright split that is harder to defend. Michelle Lewis of the Songwriters of North America put it straight: with a sound recording it is easy to tell when you have been copied, because you know exactly what Beyoncé and Drake sound like; with a composition it is more nebulous and much harder to sue over. On top of that, a proposed class action filed last month by Jason Isbell and other musicians argues that prompting a generator to produce a track evoking a specific artist's identity is a use of the performer's likeness and voice — a right the plaintiffs say belongs to the musician no matter who owns the recordings. The labels dispute the claims.

And this week, the other hand is still swinging

The cleanest illustration of how little the lawsuits have to do with protection came on Friday, 18 September. Sony and Universal filed a second, 45-page complaint against Suno in the same Boston federal court where the original case is still live. The new fight is over v6 — the very models the company built with licensed music from Warner, BMG and Believe. The labels argue that because v6 was trained in part on the outputs of Suno's older models — which, they say, were trained on unlicensed recordings — the infringement is simply carried forward.

V6 is not a fresh start; it is the fruit of the same poisoned tree.

— Sony Music and Universal Music Group, in their 18 September 2026 complaint against Suno, via Variety

Read the geometry of that position carefully, because it is the whole story. The company that is suing Suno for training on unlicensed music is simultaneously licensing its music to Udio and to Spotify. Sony and Universal are suing one AI platform while other parts of the same industry sign deals with it, and while UMG runs its own licensed AI platform with Udio on the side. The line being drawn is not between the music business and AI. It is between AI the labels control and AI they do not.

There is a genuinely surprising procedural detail buried in the response. Suno confirmed that v6 was trained in part on the 'creations' of its own users — that is, on the songs people generated with the earlier models. The company says that is a feature, not an admission: a model learning from the work its community made with it. The labels read the same sentence as evidence that v6 is downstream of the recordings their older models were trained on. It is the first time Suno has publicly put user creations inside its training account, and both sides claim it.

The counter-argument, honestly stated

The labels would say we are being uncharitable, and the case is not nothing. Licensing AI companies is, on their telling, how artists get paid at all — better a negotiated deal with a share for the rights holders than a lawsuit that ends in an injunction no one profits from. When Universal announced its settlement with Udio, its chief executive framed it as doing right by the people signed to the label.

These new agreements with Udio demonstrate our commitment to do what's right by our artists and songwriters, whether that means embracing new technologies, developing new business models, diversifying revenue streams or beyond.

— Sir Lucian Grainge, chairman and CEO of Universal Music Group, 29 October 2025

Some in the industry make the streaming comparison: this is the same fight that met Spotify, and the labels eventually found a way to make the platforms pay. One Berklee expert who has advised Suno called the deals 'a good step' and 'an inevitable step' if the industry wants to stay engaged with musicians. Suno's own line is that its platform exists 'so that more people can make new music,' trained on content 'licensed from our partners.'

The streaming comparison cuts both ways, and this is where the counter-argument stops being reassuring. Streaming grew the industry and then spent years under complaint about how little of it reached artists. The people making the comparison rarely add that the labels did not ask the artists before selling streaming either. And the terms of these new deals are undisclosed, which is the whole problem. Nobody outside the boardroom knows what artists will be paid, how many have opted in, or what the training data is worth. When the terms of the deal are secret, the claim that it is 'for the artists' is doing most of the work.

What we take from this

We review AI audio tools, so the lesson we keep drawing is the one these deals keep teaching. The 2024 lawsuits looked like the industry finally taking a stand for the people who make the music. They turned out to be a market in the making. The labels did not protect their artists from AI; they sold the training data and described the sale as protection, and the musicians have had to sue their own labels to find out what was done with their work. When a musicians' union has to sue the same companies that claimed to be defending it, you know where the money was heading all along.

The major labels are just seeing dollar signs without any regard to the long-term health or sustainability of this industry, because they're envisioning an industry without artists who can push back.

— Damon Krukowski, musician and legislative director of United Musicians & Allied Workers, in the Los Angeles Times

For anyone who makes music, or who uses an AI audio tool to make a living, the practical version is stark. Every licensing deal is a data sale, and the human performers are the people who most need a clear contract, because the label's contract already decided who owns the master and what the AI company may do with it. Opt-in, where it exists at all, is the label's opt-in, and whether it keeps a recording out of training or only out of fan-facing features is rarely clear.

The tests ahead are the ones the reporting keeps naming: whether Sony and Universal win or settle, what the first payouts actually look like, and whether an artist who says no can still be heard. On the evidence of two years, don't hold your breath for the last one. The industry that sued AI for stealing its music is now the industry selling it back — and the musicians, as usual, are the last ones asked.

We use AI tools in our research and drafting on this site. Every judgement and every recommendation is a person's.