In twelve months the record industry went from demanding takedowns of AI tracks to selling the right to make them. What actually happened and what is still missing.
In brief
- Between October and November 2025, Universal and Warner turned their lawsuits against Udio into licensing deals, with licensed platforms expected in 2026. Warner reached a similar agreement with Suno.
- The new principle is written into the deals themselves: artists and songwriters decide whether and how their name, image, likeness, voice and compositions may be used in AI-generated content and get paid for it.
- On the legal side, the US is building a federal right over digital replicas: the NO FAKES Act was advanced unanimously by the Senate Judiciary Committee on 18 June 2026.
The gap copyright never covered
Music copyright has always protected two things: the composition and the sound recording the master. It does not protect timbre. As long as replicating an artist’s voice required a human impersonator, the gap was theoretical. Once synthesis models trained on public recordings could do it, it became the industry’s central problem.
The consequence is familiar: unauthorised tracks convincingly reproducing the voices of major artists, going viral before anyone can react. The initial response was the predictable one takedown requests, one at a time, always late. Then, over a few weeks in the autumn of 2025, the strategy changed completely.
The dates that changed the picture
On 29 October 2025, Universal Music Group announced it had settled its copyright infringement litigation against the AI platform Udio and would collaborate on a new commercial music creation, consumption and streaming experience, due in 2026. The revamped Udio will feature recordings and publishing assets controlled by UMG but only from artists who choose to participate, with participating artists and songwriters to be remunerated.
Three weeks later, on 19 November, Warner Music also settled with Udio: the agreement spans recorded music and publishing and, in the company’s words, creates new revenue streams for artists and songwriters while keeping their work protected. The news landed an hour after Warner announced a partnership with Stability AI. Universal, for its part, had already formed an alliance with Stability to develop professional tools built on ethically trained models.
Then comes the passage that formalises the principle better than any statement. Under the Warner Suno agreement, artists and songwriters will have full control over whether and how their names, images, likenesses, voices and compositions are used in new AI-generated content. That is no longer a defence: it is a licence, with the switch in the artist’s hand.
In the background, the rest of the ecosystem moved the same way: all three majors licensed music to the service Klay, and Musixmatch signed AI licensing deals with Sony Music Publishing, Universal Music Publishing Group and Warner Chappell.
| The turning point | |
|---|---|
| 29 October 2025 | UMG settles with Udio, announces a licensed platform for 2026 |
| 19 November 2025 | Warner settles with Udio; announces Stability AI partnership the same day |
| Warner–Suno deal | Explicit artist and songwriter control over name, image, voice and compositions |
| All three majors | Licences to the Klay service; publishing deals with Musixmatch |
| May 2026 | Suno still in litigation with Sony, with no major-label deals |
The counter-example: who did not sign
The transition is not unanimous, and that is what separates analysis from a press release. Suno, sued alongside Udio, was still in litigation with Sony Music as of May 2026 and had signed no licensing deal with any major, having stated it intends to defend the open-platform model rather than settle. This is not a marginal player: in November 2025 the company raised $250 million at a $2.45 billion valuation.
Two opposite bets are therefore running at the same time. On one side the walled garden: licensed catalogues, opt-in artists, tracked royalties. On the other, an open platform defending in court the legitimacy of its training. Which is right will be settled by rulings and by the market, not by announcements.
How a voice bank would work
The model the majors are building has a recognisable architecture, worth describing for what it is: an industrial project under construction, not infrastructure already running at scale.
The starting point is voice as trainable data: dedicated recording sessions produce a proprietary voice model belonging to the artist, distinct from existing recordings. That model is then made available selectively to producers and companies through professional marketplaces, or to fans for authorised derivative content via subscriptions or pay-per-use. Every use should generate a share attributable to artist and label.
The two technical conditions that decide whether the model works are less eye-catching than the announcements. The first is marking: inaudible digital watermarks that make generated content recognisable and traceable to its licence. The second is attribution: knowing precisely which voice and which composition contributed to a generated track, because without that tracking, royalty splits are a promise with no accounting behind them. It is the real bottleneck of the whole sector, and nobody has solved it at scale yet.
What the law says: the digital replica right
While the industry negotiates, the law is being written and in the US at unexpected speed.
The NO FAKES Act Nurture Originals, Foster Art, and Keep Entertainment Safe was advanced unanimously by the Senate Judiciary Committee on 18 June 2026 and moved to the full Senate. If passed, it would create a federal liability framework for producing or distributing unauthorised digital replicas of a person’s voice or visual likeness, with carve-outs for First Amendment–protected uses such as news reporting, parody and criticism, and exemptions for non-profit libraries, archives and accredited educational institutions engaged in non-commercial research. The right is structured as a property right, licensable in life and transferable after death.
At state level the precedent is the Tennessee ELVIS Act, in force since 1 July 2024, which protects voice likeness from unauthorised AI replication and applies to any individual, not only celebrities. It became the national archetype, since followed by Montana, Arkansas, Washington and others.
And there is a third route, which says a lot about how unsettled the framework still is: some artists have begun registering federal trademarks covering elements of their voice and likeness Matthew McConaughey did so, followed by Taylor Swift on the theory that trademark rights offer more uniform protection than state-level publicity rights. It signals a gap rather than a solution: trademark law was not designed for AI-generated content.
In Europe the reference point is the AI Act, which imposes transparency obligations on artificially generated content and on the data used to train general-purpose models. But one technical problem no rule solves on its own remains: verifying whether a specific voice was used to train a model is, as things stand, very hard.
What this market is actually worth
Precision matters here, because this is where most coverage goes wrong. Available forecasts cover the entire voice cloning market call centres, healthcare, gaming, accessibility, dubbing not music licensing, which no analyst measures as a separate category. Grand View Research projects the global voice cloning market to reach $9.75 billion by 2030, at a compound annual growth rate of 26.1%; other research firms give different figures, between roughly $6 and $11 billion over the same horizon.
The useful lesson is not the number: it is that anyone quoting a precise figure for “the market in artists’ voices” is extrapolating. The real value of synthetic voice in music will emerge once the licensed platforms are running and the first royalty statements become public.
Frequently asked questions
Have the major labels made peace with AI? Partly. Universal and Warner turned their Udio lawsuits into licensing deals in October and November 2025, and Warner also reached an agreement with Suno. Suno, however, was still in litigation with Sony as of May 2026.
Can an artist stop their voice from being used? Under the announced deals participation is opt-in: artists decide whether to take part and how their voices may be used. Outside those deals, protection depends on national and state law.
What is the NO FAKES Act? A proposed US federal law creating a right over digital replicas of voice and likeness. The Senate Judiciary Committee advanced it unanimously on 18 June 2026; it still needs a full Senate vote.
What is the ELVIS Act? Tennessee’s law, in force since 1 July 2024, protecting name, image and voice from unauthorised AI use. It became the model for similar laws in other US states.
How can anyone tell whether a voice was used to train a model? That is the sector’s unsolved problem: after-the-fact verification is technically difficult. Which is why the debate is shifting to digital watermarking and training-data transparency requirements.
Sources
- Music Business Worldwide “Universal Music settles Udio lawsuit, strikes deal for licensed AI music platform” (October 2025)
- Music Business Worldwide “WMG settles Udio lawsuit, strikes licensing deal” (November 2025)
- Billboard UMG–Udio deal FAQ
- Copyright Alliance round-up of music industry AI licensing agreements (December 2025)
- Holland & Knight “Senate Committee Advances Bill to Protect Name, Image, Likeness and Voice Against Unauthorized AI Use” (June 2026)
- Fenwick analysis of the revised NO FAKES Act and the ELVIS Act
- Grand View Research global voice cloning market forecasts
