Sony Music and UMG Sue Suno Over v6, Alleging Earlier AI Music Infringement Was Carried Forward

Sony Music Entertainment and Universal Music Group have brought a new federal lawsuit against AI music-generation company Suno, targeting the company’s v6 family of models. The case was filed in the U.S. District Court for the District of Massachusetts, with the labels and certain affiliated imprints named as plaintiffs.

At the center of the 45-page complaint is a sharply framed allegation: Suno’s newer models cannot be treated as free of infringement merely because they were trained on material produced by older Suno models. The labels contend that Suno acknowledged v6 was trained using outputs from earlier models, and they argue those earlier models had themselves infringed the labels’ copyrighted works.

“V6 is not a fresh start; it is the fruit of the same poisoned tree,” the labels claim in the lawsuit.

Suno had not provided an immediate comment at the time of reporting. That matters because the claims described in the complaint are allegations, not findings by a court. The filing starts a legal process in which Suno can respond, challenge the labels’ account, and put forward its own explanation of how the models were built and what legal standards it believes apply.

What the labels are alleging about Suno v6

The complaint’s theory is not simply that a current product allegedly resembles protected work. Its focus, as described in the filing, is on the lineage of the training process.

Training is the process through which a machine-learning model is developed using examples or data. In this dispute, the labels’ argument is that a model trained on the outputs of a previous model may retain the legal problems they attribute to that predecessor. Put less technically, their position is that generating a new pile of training material from an allegedly infringing system does not cleanse the material’s origin.

The labels characterize that process as “laundering” infringement. That word is advocacy, rather than a legal ruling or an established description of all model-to-model training. But it captures the key dispute: whether distance from the original allegedly copyrighted material changes the analysis when an AI system’s later generation learned from an earlier generation.

This makes the suit unusually focused on an important question for generative AI development: what happens when the output of one model becomes the input for another? A company may describe a later model as newly trained, but the labels’ complaint argues that a new training run is not necessarily a genuinely clean break if its dataset includes output from an earlier system alleged to have been built on copyrighted works without authorization.

Why the phrase “model outputs” is important

AI products are often discussed as though there are only two stages: a model is trained, then users receive results. The allegation against Suno v6 turns attention to a possible third layer: synthetic training data.

Synthetic training data is material generated by a system and then used to train a later system. It can include text, images, audio, or other outputs, depending on the product. The current complaint concerns music-generation technology and says v6 learned from outputs of Suno’s earlier models.

That chain creates a practical distinction between two questions:

• What material, if any, was used to train the earlier Suno models?
• Does using the output of those earlier models to train v6 remove, preserve, or transform any alleged infringement connected to the earlier training?

The labels’ answer, as represented by their lawsuit, is clear: it preserves the problem. Their claim is that an alleged infringement cannot be eliminated by placing an additional model layer between the original works and the newer product.

Whether a court agrees will depend on arguments and evidence not established by the filing alone. The available account does not detail Suno’s expected defense, the full technical record behind v6, or any judicial assessment of the allegations. Those absences are worth keeping in mind, particularly in an area where plain-language descriptions of AI pipelines can leave out substantial technical and legal complexity.

A dispute about provenance, not just musical similarity

Provenance means the history and origin of something. In copyright disputes involving AI, provenance can become central because parties may disagree not only about an output, but also about the route by which a system learned to produce it.

Here, Sony Music and UMG are challenging the claimed separation between v6 and Suno’s earlier models. The complaint says the company has asserted that its newer models do not infringe copyrighted work. The labels’ counterargument is that the new model’s connection to prior-model outputs makes that assertion inadequate if those earlier models were trained through infringement.

That is a consequential framing for businesses developing generative systems. It suggests that a model’s compliance claims may be judged not only by its immediate dataset but by the provenance of material inside that dataset. If a training corpus contains AI-generated material, questions may follow about the data and systems that generated it.

The lawsuit therefore puts pressure on the idea that “generated by AI” automatically functions as a meaningful endpoint for rights analysis. The labels allege the opposite: generated material can still carry a disputed history.

What is known — and what is not

The reported facts establish several basics. Sony Music Entertainment and Universal Music Group filed the new case in Massachusetts federal court. The complaint is 45 pages long. It targets Suno’s v6 family of music-generation models. It includes the labels’ allegation that v6 was trained on outputs from previous Suno models, and it contends that those earlier models infringed plaintiffs’ copyrighted works. Certain imprints affiliated with the two major music companies are also plaintiffs.

There are also major limitations on what can responsibly be concluded now. The complaint is one side’s legal account. No immediate response from Suno was available. There is no reported ruling, no finding of liability, and no described outcome. It would be inaccurate to state that Suno has been found to have infringed, or that a court has accepted the labels’ account of v6’s training pipeline.

It is likewise not possible, on the supplied information, to describe v6’s consumer features, its release timing, its underlying architecture, specific songs at issue, or the remedies being sought. Those details should not be assumed merely because the lawsuit concerns an AI music product.

Why this matters beyond one version number

Version labels can sound like clean technical milestones: v5 gives way to v6, and the new release sounds separate from what came before. The plaintiffs are specifically contesting that intuition. Their “poisoned tree” language presents v6 as a continuation rather than a reset.

For creators, labels, and companies working with generative tools, the practical issue is traceability. If a system uses material generated by a prior system, stakeholders may ask whether there is a reliable record of where that upstream output came from and what rights were connected to its creation. The Suno complaint puts that concern directly before a federal court in the context of AI-generated music.

For readers following technology’s collision with creative industries, this case also shows why arguments about AI rarely stay confined to a single prompt or output. The dispute may reach backward through successive versions of a product and through the datasets used at each stage. That concern around the economics and ownership of creative production sits alongside other industry questions, including the pressures examined in this look at the reported budget behind The Blood of Dawnwalker, though the underlying legal issues are distinct.

What to watch next

The next meaningful development will be Suno’s response, if and when it is filed or publicly stated. A response could address the labels’ characterization of v6, dispute the allegation that prior models were infringing, challenge the claimed relationship between earlier outputs and v6, or raise other legal defenses.

Until then, the most accurate reading is narrow but significant: two major music companies are asking a federal court to treat training on outputs of an allegedly infringing predecessor model as insufficient to erase the alleged infringement. Suno v6 is the immediate target, but the principle being argued could matter wherever generative systems are trained on the work of earlier generative systems.