What Sony Music v. Suno Means for the Music Business

The music business is having two conversations about AI – one in public, and one in panic. The recent matter of Sony Music v. Suno is where they collide.

In my twenty-four years in entertainment law, I’ve watched this industry get repriced three times – Napster, streaming, TikTok. This is the fourth, and this time, the asset being repriced isn’t the song – it’s the right to learn from it.

The facts

In June 2024, Sony, UMG, and Warner – through the RIAA – sue Suno, a generative artificial intelligence music creation platform, in federal court in Massachusetts. Suno is alleged to have copied vast quantities of copyrighted recordings, without permission, to train its music-generation models.

Suno’s answer, in four words, was that “Learning is not infringing.” One of those positions will survive, and what’s at stake is bigger than one startup.

Music catalogues: Repertoire or training fuel?

Labels have monetised recordings the same ways for decades, through sales, streaming, sync, catalogue. AI adds a new use case where recordings are raw material. If Suno’s fair-use theory holds, AI companies get leverage to treat recorded music as input data first, licensable content second. If Sony wins, catalogue owners get to charge at the door – before training begins.

Either way, training rights are now joining sales, streaming, sync, and catalogue and becoming their own licensing category with real enterprise value. While the outcome is not settled, it’s a timely reminder to price your catalogue accordingly.

An enforcement A/B test

The industry is running a split-screen strategy where they litigate some players and license others.

Look at this very case. Warner settled with Suno and cut a licensing deal. Sony and UMG kept pressing. The majors do not want the same endgame at the same speed – some are converting litigation risk into partnership, others want a court to draw the line first.

What this means for AI companies is that there is no single industry to make peace with. There’s a patchwork of litigation, selective licensing, and strategic holdouts – all at once.

Black-box training is becoming commercially untenable.

Suno’s stated position was that it trained on music from the open internet, and this includes copyrighted material within tens of millions of audio files, per its own California AI disclosure.

That sounded abstract when the product was a toy, but it sounds different now. When labels are fingerprinting training corpora, the exposure attaches not just to startups, but to their investors, distribution partners, and enterprise customers. The cleanest provenance story wins the business-development story.

This case says “training.” It means “outputs.”

Officially, this is a training-data lawsuit. Commercially, the fear runs deeper, because these systems have the potential to not only learn from recordings, but compete with them, approximate them, and, ultimately, flood the market with style-adjacent substitutes.

The real question for rights owners is whether AI products dilute the scarcity – and the bargaining power – of recorded music itself.

Independents feel this before the doctrine settles.

The majors can litigate, settle, or hold out, but indies can’t. Distributors, managers, self-releasing artists are all deciding right now whether to work with these platforms, opt out where they can, or chase new licensing revenue.

The danger is a two-tier market. Big rights holders negotiate bespoke AI deals. Everyone else gets terms-of-service asymmetries and opaque ingestion.

For independents, the leverage is in the contract, rather than a doctrine that currently lives in the abstract.

A case that will set industry norms

A court may eventually answer the narrow question – is copying recordings into a training corpus fair use?

The business question is much wider – who decides how music enters AI systems, on what terms, with what disclosure, with what economic participation?

That’s why this is important even if it settles. The process itself is setting norms – training-data audits, fingerprint validation, documented provenance. “Responsible AI” in music is becoming a paper trail and tracking entry points, rather than analogies about how kids learn songs.

And the endgame?

If AI music becomes licensed infrastructure, labels emerge as gatekeepers and toll collectors. If Suno’s theory prevails, startups build first and license later – if at all.

The likely landing spot is the middle – licensed but concentrated. An ecosystem where only the best-capitalised AI players can afford a compliant training pipeline.

This case is significant because it goes beyond whether one startup copied songs. It goes to the whether the AI-influenced future of music gets built on permission, settlement leverage, or fait accompli.

Once the training rules are set, everything downstream gets repriced. And for that reason, it is certainly one to watch.