Sony Music and Universal Music Group have filed yet another lawsuit against AI music generator Suno, and this time the labels are leveling an unusual accusation: "model laundering." The complaint, shared with The Verge, argues that Suno's new v6 model is still built on their copyrighted recordings — even if v6 itself was never directly trained on them.
The 'Model Laundering' Argument
The core of the dispute is Suno's claim that v6 represents a clean break from its earlier models. The labels argue the opposite: that training a new model on the outputs of an infringing model does not eliminate the infringement, it launders it. For more context on this story, see our ongoing more AI stories.
"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 recordings into the tainted models, from those models into their outputs, and from those outputs into v6," the complaint states. "V6 is not a fresh start; it is the fruit of the same poisoned tree."
According to the labels, Suno's earlier models were trained on unlicensed music ripped from YouTube and other sources, and v6 inherits that taint because it was trained heavily on user outputs — the songs generated by those earlier, allegedly infringing systems.
Sony and Universal are notable holdouts in the AI music space: unlike some of their competitors, they never signed a licensing agreement with Suno, leaving the company without authorization to use their catalogs.
A New Twist: Distillation Claims
Sony's complaint goes further, alleging that Suno used distillation to train v6 — a technique in which a new model is trained to replicate the outputs of existing "teacher" models rather than learning from raw data. If v6 was distilled from predecessor models built on infringing recordings, the labels argue, the new training data changes nothing about the underlying violation.
"Even a model not directly trained on Plaintiffs' recordings is informed by, and benefits from, Suno's retained unauthorized copies," the complaint reads. In the labels' view, unless Suno truly starts from scratch, its models will continue to be built on stolen data — no matter how many times the company declares a fresh start.
Suno's Defense
Suno has publicly maintained that v6 was built differently. When the model launched, Suno's Jack Brody told The Verge it was "trained from the ground up, with a new set of data," including user data, though he offered no details beyond that.
Suno spokesperson Rachel Racusen later confirmed that the training data includes "creations" from users, saying v6 was trained on "content licensed from our partners, interactions including creations and preference signals from our community, and the accumulated learnings from our team." She did not specify whether that data included uploaded audio, or outputs derived from uploaded audio — the distinction at the heart of the labels' laundering theory.
Mounting Pressure on the AI Music Business
The new complaint adds to Suno's already crowded legal docket. The startup lost a landmark German copyright case to collecting society GEMA in August, and the major labels' unwillingness to license their catalogs leaves Suno's core training strategy exposed on two continents.
The labels' earlier US lawsuit against Suno targeted the training of its first-generation models on their recordings. The v6 complaint effectively closes off the most obvious escape route: starting over with synthetic data and user creations. If even that path is legally tainted, Suno's options narrow to licensing deals on the labels' terms, a courtroom defense of its practices, or rebuilding its training corpus from music it can prove is licensed or original.
Suno is not alone in facing this squeeze. AI music generators, image models, and text systems are all confronting the same question as the early era of scrapping the open web gives way to a legal reckoning. Some rights holders have chosen licensing over litigation, striking deals that turn their catalogs into training inputs with compensation attached. Sony and Universal, by suing twice, are betting that the courts will deliver a better outcome than any license.
The case also matters far beyond one company. Courts are increasingly being asked to decide whether AI outputs — and models trained on outputs — can carry the legal taint of their training data. The concept the labels call model laundering has obvious parallels in text and image generation, where synthetic data scraped or generated from earlier models is becoming a larger share of training sets.
If Sony and Universal's theory gains traction, it could reshape how AI companies think about provenance: not just what went into a model, but what went into the models that produced its training data. For now, the question moves to a US court, where Suno will have to argue that v6 really is the clean break it claims to be.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
