AI Music Is Being Sued and Licensed at Once

You open a music app and a track plays that sounds like a band you know, in a style you recognise, performed by nobody. It took four seconds to make and cost nothing. Somewhere, a label is suing the company that generated it and licensing its catalogue to the same company in a different deal.

That is not a contradiction anybody planned. AI music has arrived in a state of open legal war and quiet commercial cooperation at once, and the two tracks are being run by the same handful of companies.

The outcome decides who gets paid when a machine writes a song. This article lays out who is suing whom, what the licensing deals actually cover, where the damages numbers come from, and what it means for the artists at the bottom of the stack.

Start with the lawsuit, because the specific allegation is more interesting than the headline number.

What the Labels Are Actually Alleging

The original complaints against AI music platforms were straightforward: you trained on our recordings without permission. The new case is narrower and more technical.

Variety reported that Sony Music and Universal Music filed against Suno on September 18, targeting the company's v6 model specifically. The claim is that v6, which Suno presents as built on licensed material, inherited its capability from earlier models trained on thousands of unlicensed songs.

The labels call this model laundering. The argument is that you cannot clean a model by retraining it on licensed data if the outputs and training signals of the unlicensed version were used to teach the new one.

If that argument succeeds it reaches well beyond music. Distillation, where a newer model learns from an older model's outputs, is standard practice across the whole AI industry. A ruling that treats the inherited capability as derivative would change how every lab handles a training lineage, which is why the broader wave of AI lawsuits is being watched so closely.

The Same Companies Are Signing Deals

Here is where it stops looking like a normal copyright fight.

Warner Music settled out of court with Suno and signed a licensing agreement. In September, Suno released its first models built with licensed music from Warner Music Group, BMG and Believe. Rival platform Udio reached licensing agreements with Universal and Warner.

So Universal is suing Suno while licensing to Udio. Warner has licensed to both. The majors are not trying to kill AI music. They are trying to own the terms on which it exists.

That reading explains the timing. A lawsuit filed against the one large platform that has not fully settled is leverage, not abolition. Every settled deal raises the price of the unsettled one.

It also explains why the suits target the model rather than individual outputs. Winning on outputs produces takedowns. Winning on the model produces a royalty.

Where a $9 Billion Number Comes From

Damages figures in these cases sound arbitrary until you see the arithmetic behind them.

US copyright law allows statutory damages of up to $150,000 per work where infringement is found to be wilful. Plaintiffs do not have to prove what they lost. They count works.

• At the $150,000 maximum, a $9 billion claim implies roughly 60,000 individual works.
• Round Hill Music, which sued Suno separately in August in federal court in California, is seeking damages reported as potentially exceeding $1 billion, implying a catalogue in the thousands.
• A single major label catalogue runs to millions of recordings, so these counts represent a fraction of what could be asserted.

That last point is the pressure. The theoretical exposure is larger than the company is worth, which is exactly the condition that produces settlements and licensing deals rather than verdicts.

Artists are filing too. The Grammy-winning musician Jason Isbell and three others sued Suno over commercial exploitation of their identities, which is a different claim from copyright and does not settle on the same terms a label deal does.

The Fair Use Defence Just Got Weaker

Every AI company facing a training-data claim has made some version of the same argument: training is transformative, and transformative use is fair use.

At the end of September, Forbes reported that AI lost its first fair use appeal. A trial court ruling is persuasive. An appellate ruling is binding on the courts beneath it.

That shifts the negotiating position sharply. A platform that believed it had a strong defence could afford to fight. A platform whose defence has been rejected on appeal is pricing a settlement.

It does not end the question. Fair use is decided case by case on four factors, and a music case turns on different facts from a text or image case. But the direction of travel now runs against the position that training on copyrighted work is automatically permitted.

What This Means for Artists and Listeners

Three practical consequences are already visible, and none of them depends on how the Suno case ends.

Licensed AI music is becoming a product category, not a scandal. Once Warner, BMG and Believe are inside the training set by agreement, generated music stops being contraband and starts being inventory with a rights holder attached.

Payment structures remain the unanswered question. A licensing deal pays the rights holder. Whether and how that flows to the individual artist whose style a model learned is governed by contracts written long before any of this existed, which is why artist groups have been markedly less enthusiastic about these settlements than the labels announcing them.

And identity is a separate fight. Copyright protects a recording. It does not cleanly protect a voice or a style, which is why the Isbell suit is pleaded on commercial exploitation of identity rather than infringement. That body of law varies by state and is far less settled.

For listeners, the near-term effect is more catalogue and less clarity about what you are hearing. It is the same dynamic that reshaped video, where the streaming wars turned libraries into leverage.

Frequently Asked Questions

Is AI-generated music legal?

Generating and releasing it is generally legal. The disputed question is whether training a model on copyrighted recordings without a licence is lawful, which is what the current lawsuits are about. Output that closely imitates a specific artist's voice raises separate identity and publicity claims.

Why are labels suing and licensing the same technology?

Because the goal is control of the terms rather than prohibition. Warner settled with Suno and licensed its catalogue, Udio signed with Universal and Warner, and Universal is simultaneously suing Suno. Each settled deal strengthens the price the majors can demand from anyone still unlicensed.

What is model laundering?

It is the labels' term for building a new model using the outputs or training signals of an older model that was trained on unlicensed material. The argument is that the new model inherits capability derived from infringement even if its own training data is licensed.

Do artists get paid when a label licenses to an AI company?

That depends on individual recording contracts, most of which were written before generative AI existed and do not address it. This is the main reason artist groups have been far more critical of these deals than the labels signing them.

Can an AI copy a singer's voice legally?

Copyright covers the recording and the composition, not the voice itself. Claims about voice and style are usually brought under right of publicity or unfair competition law, which varies by jurisdiction. The Jason Isbell suit against Suno is pleaded on that basis rather than on copyright.

What to Watch

Two signals will tell you how this ends well before any verdict does.

The first is whether Universal settles with Suno the way Warner did. A settlement would confirm that the suits are pricing mechanisms rather than attempts to shut the technology down.

The second is whether any licensing deal publishes a flow-through rate to artists. Until a deal states what reaches the person whose work taught the model, the money is moving between corporations and the creative argument is decoration.

The technology is not going back in the box. Who collects from it is still genuinely open, and the next twelve months of filings will settle more of that than a decade of public debate has.