Suno Lawsuit: Universal and Sony Target v6 Over Model Laundering

A second lawsuit aimed at a 'clean' model

Universal Music Group and Sony Music Entertainment have sued Suno for a second time. Suno is the Cambridge, Massachusetts company whose AI generates complete songs from text prompts. The labels filed the new complaint in federal court in Boston on September 18, 2026. It names 60,202 sound recordings that they say Suno copied without permission to build its models3. The suit came nine days after Suno released v6 on September 9. Suno presented v6 as a new generation of models built with industry partners Warner Music Group, BMG and Believe7.

The main legal theory is new. The labels' first case, filed in June 2024, is still pending. The second case goes after v6 directly. According to the labels, v6 learned partly from songs users made with older Suno models and from users' choices between generated tracks. Because those older models were allegedly trained on the labels' catalogs without a license, the labels say the infringement carries forward into v69. The complaint calls this "model laundering"2. It also says Suno used distillation, training v6 to copy the outputs of earlier "teacher" models built on infringing data2. The labels sum up their position by saying v6 "is not a fresh start" but "the fruit of the same poisoned tree".

Suno says the claims are fundamentally flawed. It says v6 was trained on licensed material, community interactions and lessons its team had learned5. Suno has called the claims flawed but had not yet filed a formal answer when the case was assigned to Judge F. Dennis Saylor IV9.

Why there are two cases

The second suit exists mainly because of a scheduling decision. Through discovery, the labels used the Audible Magic fingerprinting service to find their recordings in Suno's training data. In May 2026 they asked to expand the original complaint from roughly 560 recordings to 61,02613. Suno opposed the request. It argued the labels had waited too long and should be made to file a separate lawsuit15.

On August 18, 2026, Judge Saylor denied the expansion without prejudice. His main reason was that it would disrupt the existing schedule. He also noted that the labels could still pursue valid claims and that the size of the alleged infringement is no defense3. Suno got what it asked for, in a sense. The labels did file separately, and the new case went to the same judge12.

The figures in the reporting line up closely. The new complaint covers almost all of the 61,026 works the court would not add to the first case9. Statutory damages for willful infringement can reach $150,000 per work, so several outlets put the theoretical maximum at just over $9 billion58. The labels also seek up to $2,500 for each time Suno allegedly got around YouTube's download protections4. Those numbers are a ceiling based on allegations, not a likely award. Courts have made no findings on infringement, willfulness or fair use12.

The courtroom record so far

No court has ruled on the merits in Suno's U.S. cases. Even so, the procedural rulings have mostly gone against the company. On August 18, 2026, Judge Saylor found that the labels had plausibly alleged that Suno got around YouTube's "rolling cipher" protection. He added that a final answer would need a developed factual record3.

Two days later he applied the same reasoning in Justice v. Suno, a proposed class action by independent musicians. He also rejected Suno's argument that the artists had to point to a specific infringing output to survive a motion to dismiss. The court held there is "no categorical requirement" to produce the derivative work at the pleading stage3. The complaint cited about 100 Suno outputs that allegedly replicate others' songs, and the judge found that enough for now12. Suno did win on one point: a Tennessee consumer protection claim was dismissed because the plaintiffs tried to recast it in a brief rather than an amended complaint12.

The biggest question is still undecided: whether training a music model on copyrighted recordings is fair use. An amended scheduling order set dispositive motions in the original label case for April 9, 202717. That pushed back a fair-use showdown that some observers had expected in mid-202614. Earlier explainers described a fair-use hearing as imminent, and the schedule has since moved15. Europe has moved faster. On July 31, 2026, a Munich regional court ruled for the German collecting society GEMA. It found that Suno infringed six compositions in both training and outputs and ordered disclosure. Suno said it was weighing an appeal178.

Reports disagree on how many U.S. cases Suno faces in Massachusetts. Boston.com counted about half a dozen active suits1. A review of CourtListener records found at least nine filed in the district since 2024, spread across several judges3. Among them is Lowery v. Suno, filed August 31 by Jason Isbell, David Lowery and two other musicians. That case is about identity and voiceprints rather than copyright1611. The exact count depends on what you include. Either way, the federal courthouse in Boston has become one of the main venues for U.S. law on AI and music.

Settle or sue

The new suit is unusual because Suno is no longer simply fighting the music industry. Warner, a plaintiff in 2024, settled in November 2025 and became a partner7. The terms included an undisclosed payment, Suno's purchase of Songkick from Warner, a commitment to licensed models in 2026, and download caps17. Suno announced a global partnership with BMG in August 20267. Since v6 launched, Warner, BMG and Believe have each received a share of Suno's revenue10.

Universal and Sony are using those deals against Suno. The complaint argues that Suno's later licenses show a license "was possible, just as it was from the start". In effect, it treats Suno's partnerships as proof that it chose to copy rather than pay1. The point targets the fourth fair-use factor, effect on the market. The more an active licensing market exists for training data, the harder it is to argue that unlicensed copying did no market harm.

Coverage is split on how this ends. Industry outlets describe the case as a test of whether a company can separate a new model from its predecessors' training history. They warn that a win for the labels could change how future music models are built7. Other commentators see the lawsuit mainly as leverage. Talks between Suno and the two remaining labels had reportedly reached a "hard impasse" by April 2026, centering on whether AI tracks can be shared outside Suno's platform15. One commentator puts the odds at 70% that Universal or Sony announces a settlement or licensing deal with Suno by the end of 202710. There is precedent: Universal settled with Suno's rival Udio in October 2025. That deal reportedly included per-generation royalties and coverage of future model versions17.

Our reading is that the second suit is mostly about leverage, with a real legal question attached. The "model laundering" theory pushes back against the industry's preferred fix, which is to retire a contested model and launch a licensed one. If that fix is not enough, then any AI company that trained first and licensed later carries legacy liability into every successor model. That would give the holdout labels far more bargaining power. Suno has kept raising money, including a reported $400 million round in June 2026. That suggests investors are pricing in the litigation risk rather than waiting for it to clear14.

What it means for AI policy and legal practice

The case has three implications beyond Suno.

First, provenance is now evidence. The labels grew their claim more than a hundredfold by fingerprinting Suno's training data in discovery13. One analysis argues that anyone holding audio training data should keep provenance records before a subpoena or forensic review forces the question14. For legal teams adopting AI tools, the lesson is that automated content matching has become a routine part of copyright discovery.

Second, the distillation theory reaches past music. Many AI developers train new models on synthetic data or on outputs from earlier models. If a court accepts that infringement can pass from a teacher model to a student model, the case's significance extends well beyond audio. Commentators have noted this would make rebuilding a model insufficient to clear copyright exposure7.

Third, anti-circumvention claims are surviving. The YouTube stream-ripping claim has survived dismissal in both the label and artist cases3. In the Sony–Udio dispute, a DMCA claim survived in April 202617. That matters because the labels argue that fair use cannot excuse getting around technical protections15. AI companies may therefore face liability for how they obtained data even if training itself is eventually found lawful.

Users face less risk. These suits target Suno and Udio as companies, not people who use the tools8. For now, v6 remains available and nothing in the new case has been decided9. No court dates have been set in the new suit11. U.S. courts will not give a fair-use answer before next spring's dispositive motions, so the next developments are more likely to come from the negotiations than from the courtroom.