Sony and UMG Sue Suno Again Over the Copyright Trail Behind v6

Introduction

The latest dispute between major music labels and AI music company Suno is about more than whether a generated song resembles a copyrighted recording. Sony and Universal Music Group have filed another suit against Suno, arguing that the company’s v6 model may still inherit the copyright problems associated with earlier systems through model outputs, user-created material, and distillation.

The labels’ theory: a new version is not necessarily a clean start

Sony and UMG are among the major rights holders that have not signed a licensing agreement with Suno. In the new complaint, they argue that Suno’s earlier models were trained on music allegedly obtained without authorization from YouTube and other sources. If v6 learned from the outputs of those earlier models, the labels say, the legal risk was passed forward rather than removed.

The companies call this process “model laundering.” Their argument is that the value of protected expression can travel through a chain: unauthorized recordings inform an older model, that model produces outputs, and those outputs then influence a new model. Under this theory, a model does not need to ingest the original recording directly in order to benefit from knowledge derived from it.

Sony also alleges that Suno used distillation to train v6 to reproduce the behavior of earlier “teacher” models. That would broaden the legal question beyond whether a particular recording appears in the latest training set. It could also involve whether parameters, learned behavior, or generated examples preserve and transmit the effects of unauthorized copying. These points remain allegations in an active lawsuit, not established findings by a court.

Suno has offered a partial explanation

When v6 launched, Suno executive Jack Brody told The Verge that the model was trained from the ground up with a new set of data. A company spokesperson later said that the training process involved content licensed from partners, community interactions including creations and preference signals, and accumulated learning from Suno’s team.

The company did not clarify whether those inputs included uploaded audio or outputs based on uploaded audio. That missing detail is central to the dispute. “New data” could mean a wholly separate collection of original material, but it could also describe a mixture of licensed content, user activity, filtered examples, and knowledge inherited from previous development. Without a fuller account of the pipeline, outside observers cannot easily determine where the boundary between a fresh model and a successor model lies.

Why the case matters for AI music

The lawsuit could put several questions before the courts. Does training a next-generation model on the output of an earlier system also transmit the earlier system’s copyright liabilities? Should distillation, fine-tuning, and preference learning be treated as distinct stages requiring separate documentation? And when users create material on an AI platform, what rights does the platform have to reuse that material for model improvement?

A ruling favorable to the labels could increase pressure on AI companies to maintain auditable records of training sources, permissions, filtering decisions, and model lineage. It could also shift enforcement attention away from only the final generated track and toward the way models are built and updated.

The case is still a dispute over facts and legal interpretation. Its eventual outcome could nevertheless influence how AI music companies design data pipelines—and whether outputs from one model can safely become training material for the next.

Source: The Verge AI