Sony and UMG Accuse Suno of 'Model Laundering' in New Suit

Overview

Sony Music Entertainment and Universal Music Group (UMG) have filed a new lawsuit against Suno, the prominent generative AI music company, alleging continued copyright infringement. The suit centers on Suno’s latest model, v6, which the labels claim is built upon data derived from previously infringing sources rather than being trained from scratch.

The complaint introduces a novel legal concept, accusing Suno of engaging in “model laundering.” This term describes the process of training a new AI model on the outputs of an older one that was originally trained on unlicensed copyrighted material. Sony and UMG argue that this method effectively sanitizes stolen intellectual property, allowing Suno to continue exploiting their artists’ work without direct authorization.

Unlike some competitors in the AI music space, Sony and UMG did not sign licensing agreements with Suno. The lawsuit represents the latest escalation in a long-running legal battle between the tech giant and the world’s largest music publishers, raising significant questions about the future of training data in generative AI.

The Core Allegation: Model Laundering

The central thrust of the new complaint is that Suno’s v6 model is not a clean slate, as the company has publicly claimed. Instead, Sony and UMG assert that v6 was trained using outputs from Suno’s previous models, which themselves were built on unauthorized copies of music scraped from YouTube and other online sources.

“Training a ‘new’ model on the outputs of an infringing model does not eliminate the infringement; it launders it,” the complaint states. “V6 is not a fresh start; it is the fruit of the same poisoned tree.”

The labels argue that this process passes the value of their copyrighted expression through multiple layers. First, the value is embedded in the initial infringing models. It then moves into the intermediate outputs generated by users, and finally into v6. By claiming “model laundering,” Sony and UMG are attempting to close a legal loophole where AI companies might argue they are no longer directly infringing because they are only using “transformative” user-generated outputs.

Distillation and Teacher Models

Beyond the concept of laundering, the lawsuit specifically accuses Suno of using knowledge distillation to train v6. In this process, a smaller or newer model (the “student”) is trained to replicate the behavior and outputs of a larger, pre-existing model (the “teacher”).

Sony alleges that Suno’s previous models acted as teachers, containing the proprietary patterns and styles of the labels’ artists. By forcing v6 to mimic these outputs, the argument goes, Suno effectively retained and utilized the infringing data, even if the raw audio files were deleted.

The complaint emphasizes that even if a model is not directly trained on the original recordings, it is still “informed by, and benefits from, Suno’s retained unauthorized copies.” This suggests that the labels view any derivative use of training data as a continuation of the original infringement.

Suno’s Defense and Data Claims

Suno has consistently maintained that its training processes have evolved. When v6 launched, Jack Brody, Suno’s chief business officer, told The Verge that the model was “trained from the ground up, with a new set of data.” He noted that this included “user data,” but declined to provide further specifics at the time.

Subsequently, Suno confirmed to Engadget that its training data includes user “creations,” but refused to elaborate on the volume or origin of that data when asked for comment regarding the lawsuit. This lack of transparency has fueled the labels’ suspicions that Suno is obscuring the provenance of its training datasets.

The Licensing Disconnect

A critical factor in this legal dispute is the absence of a commercial agreement. While some AI companies have entered into licensing deals with major music publishers to secure the right to use their catalogs, Sony and UMG remain holdouts. They have not signed such agreements with Suno, meaning any use of their copyrighted material is considered unauthorized under current law.

This standoff highlights a broader tension in the AI industry. As courts grapple with whether training on copyrighted works constitutes fair use, major rights holders are increasingly turning to litigation to establish precedents. The “model laundering” allegation is particularly significant because it challenges the technical feasibility of “clean room” training strategies often touted by AI developers.

Industry Implications

The outcome of this lawsuit could have far-reaching consequences for the generative AI sector. If courts accept the argument that training new models on the outputs of old, infringing models constitutes continued infringement, it could force AI companies to either:

Purge all historical training data, including user outputs, which may degrade model quality.
Negotiate broader licensing deals, potentially increasing costs for AI developers.
Re-evaluate distillation techniques, as these may become legally risky if the “teacher” models contained unauthorized data.

This case also intersects with other recent legal actions in the music industry. For instance, UMG has separately sued DistroKid for alleged deceptive trade practices and copyright infringement, while Sony has been involved in broader disputes over AI-generated content. These parallel actions suggest a coordinated effort by major labels to tighten control over how their intellectual property is used in AI development.

Precedent for Future Lawsuits

Legal experts note that the “model laundering” theory is relatively new in copyright litigation. If successful, it could expand the scope of liability for AI companies, making it harder for them to distance themselves from past infringement claims. Conversely, if Suno prevails, it may validate the use of iterative training methods that rely on synthesized or user-generated data as a “clean” alternative to direct scraping.

As the case proceeds, it will likely draw attention from policymakers and regulators concerned about the balance between technological innovation and intellectual property rights. The details revealed during discovery—particularly regarding Suno’s data pipelines and v6’s training architecture—will be critical in determining the next phase of this high-stakes legal battle.