Sony and UMG Sue Suno Again Over v6 'Model Laundering'

Sony Music and Universal Music Group have filed a second lawsuit against AI music startup Suno, arguing that the company's newest v6 model infringes their copyrights even though it was built differently from the versions at the center of their original case.

The new complaint, shared with The Verge, centers on a single contested claim: that Suno trained v6 on the outputs of its own earlier models, which were themselves trained on unlicensed music scraped from YouTube and other sources. The labels call this practice "model laundering."

Their filing puts it bluntly: "Training a 'new' model on the outputs of an infringing model does not eliminate the infringement; it launders it, passing the value of Plaintiffs' expression from the copied reco…" The quote, truncated in the material shared with The Verge, frames the core legal theory of the case — that derivative training does not launder copyright liability.

Why the labels went back to court

The first lawsuit against Suno, brought by the major labels, alleged that the company trained its models on copyrighted recordings without licenses. Since then, the contours of the dispute have shifted. Suno has admitted in a court filing that it scraped audio from YouTube. It has also released v6, which the company presented as its first AI music model developed with help from the record industry.

Not every label signed on. Sony and UMG are notable holdouts that did not enter a licensing agreement with Suno, and the new suit makes clear that the absence of a deal has not softened their position.

The complaint argues that v6 is, in the labels' words, "the fruit of the same poisoned tree." Even if v6's training pipeline included user-generated outputs rather than raw copyrighted recordings, Sony and UMG contend that those outputs ultimately derive from infringing training data — and that the resulting model still carries the value of their artists' expression.

What is at stake

The case tests a question the AI industry has so far avoided answering in court: whether retraining a model on its own outputs — or on user generations produced by earlier versions — resets legal exposure. If the courts accept the labels' "model laundering" theory, AI companies would face a durable liability problem. Each new model generation could inherit the alleged infringement of its predecessors, no matter how the training process is restructured.

That has implications well beyond music. Text, image, and video model developers routinely release successive versions trained on a mix of licensed data, web data, and, increasingly, synthetic outputs from prior models. A ruling that treats synthetic data derived from allegedly infringing sources as tainted would complicate that entire pipeline.

For Suno, the timing matters commercially. The company has positioned v6 as a cleaner generation of its technology, developed with record industry participation. A second suit from two of the three major labels undercuts that narrative and keeps the company tied to litigation even as it seeks to normalize its standing with rights holders.

The holdouts define the fight

Suno's path to legitimacy has run through licensing. Some labels have chosen to settle their differences with the company through deals, and Suno marketed v6 as the product of that cooperation. Sony and UMG have chosen the opposite route. Their refusal to license, followed by a second lawsuit, signals that they see the underlying training practices — not just the absence of payments — as the problem.

The "model laundering" framing makes that explicit. The labels are not merely arguing that Suno owes them for past scraping. They are arguing that the infringement propagates forward into every model built on top of the old ones, and that no amount of architectural or procedural change in v6 cures it.

Suno has previously acknowledged, in a court filing, that it scraped audio from YouTube — an admission that removed one layer of dispute about the underlying facts and moved the fight toward the legal question of whether such scraping constitutes infringement in the first place.

What comes next

The second suit ensures that the question of how AI training data propagates through model generations will be litigated, not settled quietly. If Sony and UMG prevail on the "model laundering" theory, AI developers across sectors will face a new constraint: synthetic data generated by an infringing model could itself become evidence of infringement. If Suno prevails, retraining on user outputs could become a standard defense against claims tied to earlier training runs.

Either outcome will shape how the next generation of AI models gets built — and who gets paid for the music, text, and images they learn from.