Sony Music Entertainment, Universal Music Group and affiliated labels have filed a new copyright lawsuit against Suno in the U.S. District Court for the District of Massachusetts, accusing the AI music company of carrying alleged infringement from earlier systems into its newer v6 models. The complaint, filed September 18, 2026, frames the dispute around what the labels call “model laundering”: the claim that a model trained on outputs or signals from an allegedly infringing predecessor remains legally tainted.
The claims over Suno’s v6 model
The new case, UMG Recordings, Inc. et al. v. Suno, Inc., is separate from the labels’ original 2024 litigation against Suno. In the newly filed complaint, the plaintiffs allege that Suno copied their recordings without authorization to train earlier models, then used outputs and preference data from those systems in developing v6. Those assertions are allegations in a complaint, not findings by a court, and Suno has not been found liable on the new claims.
The filing says v6 was released on September 9 and alleges that Suno trained it using “user interactions” with prior versions of its service. According to the labels, those interactions include generated outputs and preference signals derived from older models. The central premise is that if those older models were built with unlicensed recordings, their generated material can transmit the value of the underlying music into a later system.
In a key passage, the complaint argues that training a “new” model on the outputs of an infringing one does not eliminate infringement, but instead passes the alleged value of the plaintiffs’ expression from copied recordings, through the prior models and their outputs, into v6. The labels also characterize v6 as not being a clean break from its predecessors. That language states the plaintiffs’ legal theory; it does not establish that synthetic outputs necessarily retain protectable expression from any particular recording, or that a court will accept the theory.
Alleged knowledge distillation
The complaint also alleges, “on information and belief,” that Suno used knowledge distillation in developing v6. In machine-learning terms, distillation generally describes training a newer, or “student,” model to reproduce capabilities or behavior learned by an earlier “teacher” model. The labels contend that the prior Suno models were the teachers in this instance, and that their alleged exposure to unlicensed recordings was therefore transmitted to v6. Suno has not publicly confirmed that characterization.
Suno has disputed the premise of the new suit. In a statement reported by Music Business Worldwide, the company said the claims are “fundamentally flawed on both the facts and the law.” It said v6 was trained on content licensed from partners, community interactions including creations and preference signals, and the company’s accumulated learnings. The company did not publicly detail the relative share of those sources in the v6 training mixture, nor whether particular categories of user creations included outputs from earlier Suno models.
That distinction is at the center of the case. At v6’s launch, Suno described the new generation as developed with industry partners including Warner Music Group, BMG and Believe, and said it would retire previous models as it moved users to v6. The company’s position is that v6 represents a new model generation built with licensed-partner participation; the complaint instead asks whether its development remained legally or technically connected to predecessors through user creations, preference data or model-transfer methods.
What the plaintiffs are seeking
The plaintiffs’ new complaint identifies 60,202 sound recordings that they allege Suno copied without a license. It seeks damages and injunctive relief, among other remedies. The filing also says the original action involved a much smaller illustrative set of works and that the plaintiffs brought the additional claims in a new case after the court declined to add the larger group of recordings to the earlier litigation on the existing schedule.
The new lawsuit therefore does not turn only on the more familiar question of what recordings entered a model’s training corpus directly. It also presents a downstream-lineage argument: whether a developer can rely on model outputs or behavioral signals produced by an earlier system when the underlying training of that system is itself challenged. The labels’ theory depends on persuading the court that those intermediate materials do not sever the alleged connection to the original recordings.
Suno’s response highlights the competing account. The company says its v6 work involved licensed partner content, community participation and its accumulated technical learning. That characterization does not itself resolve how a court will evaluate the provenance of user-created material or preference data. Nor does the complaint establish, at this preliminary stage, that any particular v6 output contains protected expression from a particular plaintiff recording. Those issues are likely to require evidence about both the model-development process and the legal significance of the data at issue.
The licensing backdrop
The dispute arrives as licensing has become a prominent part of the music industry’s approach to generative AI. Warner Music Group announced a partnership with Suno in November 2025 that settled prior litigation between those parties and described a plan to develop licensed AI music offerings. Warner said the arrangement was intended to compensate and protect artists, songwriters and other rightsholders. That agreement does not resolve the separate claims brought by Sony and Universal, but it illustrates the commercial backdrop against which the new lawsuit will be contested.
The broader industry context also shows why the distinction between a licensed dataset and a model’s full development history may become more important. Suno has presented v6 as a model family developed with music-industry partners, while Sony and Universal contend that the pathway into a new system matters as much as the material directly supplied to it. The case could make the documentation of model lineage—covering training material, generated outputs, feedback systems and transfers of learned behavior—a more central issue in AI copyright disputes.
What the case could test
For AI developers, the case could test a question that extends beyond music generation: whether synthetic data, user-generated outputs and model-to-model transfer can reduce exposure associated with an earlier training corpus. The labels’ position is that downstream use of outputs from an allegedly infringing model cannot sever the connection to the original recordings. Suno’s response points instead to licensed partner content, community participation and the company’s own accumulated work. No court has yet resolved that disagreement, and the lawsuit should not be read as a ruling on the legality of recursive training, synthetic-data pipelines or knowledge distillation more broadly.
The case number listed on the filed complaint is 1:26-cv-14275. Suno’s public response addresses the core allegations, but the company has not publicly provided a fuller technical accounting of v6’s training data. The next stages of the litigation are likely to put unusual pressure on both sides’ descriptions of model lineage: not simply what data entered a model directly, but how prior models, their outputs and the feedback surrounding them were used to create the next one.
