The Labels Stopped Suing Long Enough to Sign

On September 10, two things happened in AI music that would have been unthinkable eighteen months ago.

Universal Music Group announced a multi-year deal with ElevenLabs to co-develop an AI platform where fans can remix, mash up, and reinterpret UMG catalog. Artists must opt in before their work is available. ElevenLabs — valued at $11 billion, more than twice Suno's valuation — is now UMG's platform partner.

Suno shipped v6, a three-model family (flagship v6, v6-wild, free v6-mini), 5× faster than v5.5, trained on licensed partner data from Warner Music Group, BMG, and Believe, with revenue sharing for artists and labels.

And UMG and Sony Music are still in court against Suno over alleged copyright infringement.

That is not confusion. That is three different strategies executing simultaneously, and the shape of them tells you how the AI music fight actually ends.

The litigation was never the endgame

When the majors sued Suno and Udio in June 2024, the framing was existential — unlicensed training on the recorded-music catalog, statutory damages per work, an industry defending itself.

What the suits actually produced was leverage. UMG settled with Udio and is now co-developing an AI platform with it. Warner settled with Suno and is now the licensed-data partner behind v6. The lawsuits established that training on catalog without permission carries legal risk, which established a price, which turned an existential threat into a licensing category.

This is the recorded-music industry's oldest move, run again. Napster was litigated into oblivion and the settlement architecture became iTunes. YouTube was sued and the settlement architecture became Content ID and a revenue split that now pays out billions. The majors do not win by killing the technology. They win by making the technology pay a toll, and the toll rate is set in court.

UMG suing Suno while partnering with ElevenLabs is not inconsistent. It is negotiating with one and demonstrating to the other what non-cooperation costs.

Opt-in is the term that matters

Buried in the UMG-ElevenLabs announcement is the clause with the most downstream consequence: artists must opt in before their work appears on the platform.

Every previous licensing regime in recorded music has been label-decides. The label owns the master, the label licenses the master, the artist gets whatever the contract says. Streaming worked this way. Sampling clearance works this way. Sync works this way.

Opt-in at the artist level is a structural concession, and it exists because the thing being licensed is different in kind. A remix platform does not just use a recording — it generates new work in the artist's voice and style. UMG can license a master. It cannot credibly license an artist's identity over their objection without creating a public-relations problem and, in several jurisdictions, a right-of-publicity problem.

The practical consequence: catalog depth on these platforms will be uneven and driven by artist sentiment, not by label leverage. Expect estates and legacy acts to participate readily — there is upside and no career risk — and expect a meaningful share of active artists to decline.

Suno's position is the strange one

Suno is now simultaneously: licensed by Warner, BMG, and Believe; sued by UMG and Sony; and shipping a free tier.

That means v6 was trained on a partial catalog. Warner is roughly 16% of global recorded music. UMG is close to a third. A model with licensed Warner data and no UMG data has a real hole in it, and "we cannot generate anything in the style of a third of popular music" is a product limitation users will notice even if they cannot name it.

Suno's calculation appears to be that shipping a licensed, revenue-sharing product creates settlement pressure on the remaining plaintiffs — a demonstration that the cooperative model works and pays, aimed at UMG's and Sony's boards rather than their lawyers. Given that UMG already settled with Udio on those terms, the calculation is not unreasonable.

The free v6-mini tier is the aggressive part. Giving away generation capacity while two majors are actively litigating is a bet that user scale is worth more than litigation optics. It may be. It is also the kind of decision that shows up in a damages filing.

What this means if you make things

Three practical shifts.

Licensed generation is about to become the default expectation for commercial use. Anyone putting AI-generated music into a monetized video, an ad, or a game has been carrying unquantified copyright exposure. Platforms with actual label deals and revenue sharing remove that exposure. That is worth paying for, and the free-and-unlicensed tier of this market is going to get squeezed out of professional workflows.

Revenue share means AI music becomes a payout line, not just a cost line. Artists on Suno's licensed partners and on UMG's opt-in platform now have a mechanism to earn from derivative generation. Small per-use amounts at very high volume is exactly the shape of streaming economics, with the same concern: the split will favor catalog owners over performers unless contracts are renegotiated.

Voice and music are converging into one licensing problem. ElevenLabs got here from text-to-speech. UMG's deal covers "personalized vocal experiences." The distinction between synthesizing a voice and synthesizing a song is collapsing, and the rights framework being built for one will govern both.

The open question

Nobody has said what happens to the unlicensed generation already in the world — millions of tracks made on earlier model versions trained on catalog nobody cleared, now sitting on YouTube, Spotify, and in commercial projects.

Every settlement so far has been forward-looking. The retroactive question is the expensive one, and it is still unanswered.