Sony and Universal File New Lawsuit Claiming Suno’s v6 Model Inherits Copyright Infringement

On September 25, 2026, Sony Music Entertainment and Universal Music Group filed a fresh copyright complaint against Suno, the AI‑music startup, alleging that its newly released v6 generative‑music model continues to draw on recordings that the two labels never licensed.

The complaint claims that Suno’s v6 version is not a clean‑room effort but rather an evolution of earlier models that were trained on large collections of copyrighted audio harvested without permission.

Background on Licensing Agreements

According to the filing, Suno previously negotiated licensing deals with Warner Music Group, BMG and Believe, deliberately leaving Sony and Universal out of those contracts. The plaintiffs argue that this selective licensing demonstrates an intent to bypass the two major labels while still benefiting from their catalogues.

The omission, they say, forms the factual backbone of the case because it indicates that Suno’s training data included unlicensed material from the two companies’ holdings.

Allegations About Earlier Models

The complaint states that Suno’s first‑generation models were built on a corpus of recordings scraped from YouTube and other online platforms without the consent of rights holders, and that those recordings constitute the “source recordings” at the root of the alleged infringement.

It further alleges that the v6 model was not trained from scratch; instead, it was exposed to user‑generated creations that themselves were produced by the earlier, allegedly infringing models, creating a chain of copying that the plaintiffs say cannot be broken by merely swapping out the raw dataset.

Model Laundering and the Poisoned‑Tree Analogy

The filing introduces the term “model laundering” to describe Suno’s alleged practice of trying to cleanse its data pipeline by re‑training on derivative outputs, likening the process to a poisoned‑tree scenario: if the roots are contaminated, any fruit that grows will also be tainted.

Suno co‑founder Jack Brody responded that v6 was trained “from the ground up” on a new dataset that incorporated user data, confirming the presence of user‑generated content but refusing to disclose the provenance of those files.

Technical Claims About Distillation

Sony also alleges that Suno employed a technique called distillation, where a larger “teacher” model transfers its behavior to a smaller “student” model. The plaintiffs argue that if the teacher model was trained on unlicensed recordings, the distilled student model—v6—will inevitably reproduce the same infringing behavior.

The legal theory advanced by the labels is that indirect training does not sever the link to the allegedly unauthorized source recordings; even if v6 never directly saw the original tracks, exposure to the outputs of earlier models is sufficient to establish liability under current copyright doctrine.

• Suno’s earlier models trained on unlicensed YouTube recordings
• User creations derived from those models fed into v6 training
• Distillation process transferred behavior from teacher to student model
• Complaint frames the chain as a poisoned tree that remains tainted

It is important to note that the allegations remain claims in a complaint. No court has yet ruled on whether Suno’s v6 model actually infringes any copyrights, and the lawsuit’s arguments have not been adjudicated.

If a court ultimately accepts Sony and Universal’s theory, the decision could set a precedent for how AI‑generated content is treated when it is trained on outputs of earlier, potentially infringing models, broadening liability beyond direct copying to include indirect, derivative training pipelines.

For firms that rely on AI‑generated music—whether for advertising, video production, or streaming services—the lawsuit signals a need to scrutinize not only the raw data used for training but also the lineage of any model they adopt, demanding transparent documentation of a model’s training history.

In Los Angeles, industry observers warned that the case could prompt a wave of contractual revisions, with music publishers likely to require AI developers to certify that every layer of their training pipeline is free from unlicensed material before granting access to their catalogs.