Sony and UMG sue Suno a second time — filed September 18 in Boston federal court — and the complaint's theory is 'model laundering': v6 is built on outputs of models trained on their recordings, the labels allege

“Sony and Universal's new Suno lawsuit says its shiny v6 model isn't a fresh start — it's, in the complaint's words, the fruit of the same poisoned tree.”

Usable claims

• On Friday, September 18, 2026 (calendar-verified), Universal Music Group and Sony Music Entertainment sued Suno for a second time; Music Business Worldwide reported the filing that day, and The Verge's write-up followed on Friday, September 25, 2026. Per MBW (Tim Ingham, datePublished 2026-09-18T17:12:42+00:00): "The joint complaint, filed on Friday (September 18) in Boston federal court, accuses the AI music company of copying 60,202 of the labels' sound recordings, without a license, and using them to build the models that run its music generation service." MBW's own arithmetic on damages: "US copyright law sets damages for willful infringement at up to USD $150,000 per work. At that ceiling, the 60,202 recordings would carry a theoretical maximum of just over $9 billion." — MBW's calculation, not a figure from the complaint. The labels are also asking for up to $2,500 for each act of circumventing YouTube's anti-downloading technology — the maximum US law allows on that separate claim — which, applied once to each of the 60,202 recordings, "would add roughly $150 million." Again MBW's arithmetic. The labels are seeking a court order stopping Suno from any further use of their recordings, and have demanded a jury trial. The new case was triggered by a recent ruling in the first case (originally filed in June 2024): per MBW, "Judge F. Dennis Saylor IV refused that request on August 18, ruling that the extra works would delay a case he wants decided on fair use, and noting that the labels could bring them as a separate suit." The request had been to add 61,026 recordings to the first case, which covers 560 works. The complaint's central new allegation concerns v6, the model suite Suno launched on September 9. Per MBW: "Suno says v6 was built in partnership with Warner Music Group, BMG, and Believe." Suno's Chief Product Officer Jack Brody told MBW at launch that "v6 was trained entirely from scratch, from the ground up," on data that "doesn't include data from Universal or Sony." The labels argue that does not wipe out liability because, they allege, v6 was built on the output of models trained on their recordings. The complaint states, in MBW's rendering: "Training a 'new' model on the outputs of an infringing model does not eliminate the infringement; it launders it, passing the value of Plaintiffs' expression from the copied recordings into the tainted models, from those models into their outputs, and from those outputs into v6," and: "…v6 is not a fresh start; it is the fruit of the same poisoned tree." Per MBW, the filing alleges Suno trained v6 using 'user interactions' with previous iterations — which the complaint describes as outputs of, and preference data derived from, the prior models (Suno generates two tracks per prompt; the preference data is which one the user picks). The filing also alleges knowledge distillation, "whereby a new 'student' model is trained to reproduce the learned behavior of a predecessor 'teacher' model." Here, the teachers are Suno's prior models. The filing names v4, v4.5, v4.5+, v5, v4.5-all, and v5.5 as trained on the same copied corpus, and states: "Thus, every model Suno has ever released – from its earliest versions through v6 – is the product of, and continues to exploit, Suno's unauthorized copying of the Copyrighted Recordings," per the complaint as MBW renders it. The complaint also turns Suno's own deals against it: per MBW, the company settled with Warner Music Group in November 2025, signed BMG on August 12, and announced a partnership with Believe on September 8, and the filing reads: "Three agreements with three major rightsholders in less than a year confirm that a functioning market exists for licensing sound recordings to train generative AI models and that Suno itself recognizes that using copyrighted sound recordings for this purpose requires permission," per MBW. The complaint quotes Brody saying the revenue share with partners is "not in exchange for training" and that the deals are "not really about the data". It calls that characterization "self-serving." On market harm, the complaint points to Deezer, which said in July 2026 that more than half of new tracks arriving on its platform each day were AI-generated, at around 90,000 a day. Per MBW, the labels pinpointed their works inside Suno's training data using audio fingerprinting company Audible Magic, during discovery in the first case. Suno argues in its first-case defense that training its models on copyrighted recordings is "quintessential fair use." It admitted in a September 1 filing that it obtained audio from YouTube using the tool YT-DLP, and fact discovery in the first case closes on September 30. Per The Verge (September 25): "Sony and UMG are notable holdouts who did not sign a licensing agreement with Suno." The Verge's rendering of the complaint: "Training a 'new' model on the outputs of an infringing model does not eliminate the infringement; it launders it, passing the value of Plaintiffs' expression from the copied recordings into the tainted models, from those models into their outputs, and from those outputs into v6 … V6 is not a fresh start; it is the fruit of the same poisoned tree." When v6 launched, Suno's Jack Brody told The Verge it was "trained from the ground up, with a new set of data," including "user data," but did not offer details beyond that. Per The Verge: "Suno later confirmed to Engadget that data includes 'creations' from users, but would not elaborate further when reached for comment." Per The Verge: "Sony further alleges that Suno used distillation to train v6. It claims that it was trained to replicate the results of its previous, 'teacher' models, which were created using infringing data." And the complaint says: "even a model not directly trained on Plaintiffs' recordings is informed by, and benefits from, Suno's retained unauthorized copies."