Suno Copyright Lawsuit Survives as Judge Rejects Output Demand

A procedural loss with real weight

Suno, one of the best-known AI song generators, has failed to get an independent artists' class action thrown out at the pleading stage. In Justice v. Suno, Inc., Chief Judge F. Dennis Saylor IV of the U.S. District Court in Massachusetts rejected Suno's argument that the musicians had to point to a specific Suno-generated track copying their songs before their derivative-work claim could go forward.11 The plaintiffs are independent artists and rights-holding companies. They say Suno's generator produces music derived from their copyrighted compositions and recordings, which they allege the company pulled from YouTube.11

The reporting is unclear on one point: the date. Legal trade publications carried the story in late September. A Boston legal outlet that read the full memorandum dates the ruling to August 20, 2026, and says Massachusetts Lawyers Weekly first reported it on September 28.8 A music-creator rights tracker also puts the refusal to dismiss on August 20. The likeliest explanation is that the decision was issued in August and only drew wide notice weeks later.

What the judge actually decided

The decision is nine pages long and splits three ways.11 The derivative-work claim under 17 U.S.C. §106(2) survived. Saylor relied on earlier federal decisions from California and Pennsylvania and found no "categorical requirement" that a plaintiff attach the allegedly infringing work at this stage.11 The artists' claim under the Digital Millennium Copyright Act's anti-circumvention provision, §1201(a)(1), also survived.2 Their claim under the Tennessee Consumer Protection Act was dismissed.2

The Tennessee dismissal came about through a pleading mistake. The artists said in their opposition brief that citing the Tennessee statute was "inadvertent" and that they had meant to plead common-law unfair competition. The court ruled that a brief cannot be used to amend a complaint.8 On the DMCA count, Saylor followed his own August 18 ruling in the major labels' case. There he held that allegations Suno got around YouTube's "rolling cipher" protection stated a plausible claim, while saying the final answer would need a developed factual record.8

The core of the ruling is about circumstantial evidence. The amended complaint offered about 100 examples of Suno outputs that allegedly imitate or replicate existing songs or artist identifiers. It quoted remarks attributed to Suno's CEO about the model's ability to reproduce real artists' vocals and songs. It also alleged that Suno built its model to produce music that mirrors its training material.11 From those allegations, Saylor found it plausible that the system had also produced outputs substantially similar to these plaintiffs' works. He said proof could wait for discovery.11

Two readings of the same complaint

The briefing shows how differently each side sees AI liability. Suno's lawyer, Brittany N. Lovejoy of Latham & Watkins, said the 66-page amended complaint never cited a single Suno output that sounded like the plaintiffs' own songs. She described the examples as third-party creations resembling other people's works.11 The plaintiffs' lawyer, Krystle Delgado, said Suno wanted a rule no court had adopted: that AI copies escape liability unless the infringing files are attached to the complaint.11

The judge sided with Delgado's view of the pleading standard. That should not be confused with a finding that Suno infringed anyone's work. A motion to dismiss treats the complaint's allegations as true and asks only whether they state a claim. The court has made no finding on infringement or fair use.8 Travis J. McDermott of Partridge, Snow & Hahn, who is not involved in the case, said the result was correct for a ruling at the motion-to-dismiss stage.12

Even so, the ruling matters more than its procedural posture suggests. The "show us the infringing output" argument is becoming a standard defense across AI cases. This month Anthropic moved to dismiss BMG's U.S. copyright claims over musical compositions, arguing that BMG cannot identify any infringing outputs from its Claude models. Anthropic also pointed to stronger guardrails in its newer models.1618 Saylor's reasoning does not bind the court hearing that case. Still, it gives plaintiffs a recent federal decision holding that pattern evidence and design allegations can carry a claim into discovery.

The courts are not uniform, though. A case tracker covering the parallel litigation against Udio shows the court there dismissing an output-based infringement count with leave to amend, while letting a DMCA count proceed.14 The Ninth Circuit, in the GitHub Copilot appeal, affirmed dismissal of claims under DMCA §1202. It reasoned that a tool that creates new works cannot "remove" copyright management information those works never contained. It expressly declined to say whether output similarity could support an ordinary infringement claim.17 Taken together, output claims appear to survive when plaintiffs bring concrete examples of imitation, and to struggle when they do not.

Boston becomes the center of AI music litigation

The Justice ruling is one of several Suno cases. Universal and Sony's 2024 suit is before the same judge. On August 18 Saylor refused to let the labels add more than 61,000 recordings to that case, mainly because of the effect on the schedule. He noted that the size of the alleged infringement is no defense and that the labels could sue separately.8 They did so on September 18, filing a second complaint over 60,202 recordings. It seeks statutory damages of up to $150,000 per work for willful infringement and has also been assigned to Saylor.8 Court records show at least nine suits against Suno filed in the District of Massachusetts since 2024.8 Separately, an independent-artist class action filed in Illinois has been ordered transferred to the Southern District of New York.14

What discovery has turned up raises the stakes. According to the labels, Suno admitted training on "tens of millions" of recordings that presumably included theirs, and a forensic analysis found millions of their tracks.8 Summary judgment briefing, where fair use will be argued, is reportedly proposed for June 2027, so a decision on whether training is lawful probably won't come before then.

The view from Munich, and from the appeals courts

Outside the U.S., Suno has already lost. On July 31, the Munich Regional Court ruled for the German collecting society GEMA. It ordered injunctive relief, disclosure of revenue and damages over six works, including "Rasputin," "Daddy Cool" and "Mambo No. 5."16 The court found the songs were "reproducibly contained" in Suno's v3.5 and v4 models. It said the text-and-data-mining exception did not apply because the works were retained in the model rather than merely analyzed.6 Coverage agrees on the main finding, though outlets frame it differently. Variety said the decision's specifics were not made public.4 JUVE Patent and Music Ally quoted the court's own press release in detail.16

The Munich court also applied U.S. law to the copying Suno did during training in America. It held that fair use did not protect Suno and distinguished the Bartz and Kadrey rulings, where U.S. judges found AI training to be fair use largely because the training works did not appear in outputs.6 A German court's view of U.S. fair use does not bind U.S. courts. Its reasoning lines up with the Justice plaintiffs' theory, though: in both, the alleged problem is outputs that copy what went in. Suno says the ruling misunderstands its technology and is considering an appeal.7

On September 29, a federal appeals court issued the first appellate ruling on AI training and fair use, ruling against ROSS Intelligence. It found market harm where content was copied to build a competing tool.10 Forbes's analysis argues that this precedent, together with Munich's memorization finding, puts Suno's U.S. fair use defense under real pressure.10 That case involved a different technology and market, but the commentary is right that market substitution is the factor plaintiffs will press hardest.

Policy pressure builds

McDermott described the current state of the law as a "Wild West." He said AI models disrupt long-settled IP rules by ingesting data and producing variations of it.11 Without action from Congress, he warned, results will be "all over the map."12 Proposals are pending. The CLEAR Act, S.3813, would require AI developers to disclose their training data to the U.S. Copyright Office before releasing a model. The Copyright Office is also taking public comment on streaming fraud and AI-generated music through November 23.

In practice, the industry is moving faster than legislators. Suno says its V6 model was trained only on licensed catalog and its own users' creations. The labels' second suit argues V6 still carries over what the earlier models learned. GEMA's general counsel has asked European lawmakers to keep AI companies from relocating to jurisdictions with weaker copyright enforcement.7

The bottom line

Justice v. Suno does not decide whether AI music generators infringe copyright. It is significant because it removes a shortcut defendants hoped to use. If the ruling holds and other courts follow it, a plaintiff who can show a system's tendency to imitate does not need to locate a matching copy of their own song before getting discovery. In discovery, the training-data records that Suno's opponents say it concealed may be opened up.8 For AI developers, the realistic choices are licensing, guardrails that hold up to forensic testing, or years of litigation where the evidence is examined in open court.