Sony Music and Universal Music Group say Suno's new v6 model didn't wash away the infringement of its old ones, it just carried the problem into a cleaner-looking package.
On September 18, Sony Music Entertainment and Universal Music Group filed a 45-page lawsuit against Suno in the U.S. District Court for the District of Massachusetts. Their argument is simple: the AI music startup's newly licensed v6 model family is still built on stolen material. The complaint names 60,202 specific recordings. It landed nine days after Suno launched v6 to the public.
The labels' theory is blunt. According to Variety, the complaint says v6 was trained partly on outputs generated by Suno's earlier, unlicensed models, so whatever legal cleanup Suno did on the front end never touched the back end. "Training a 'new' model on the outputs of an infringing model does not eliminate the infringement, it launders it," the labels wrote. Variety also reported the complaint's sharper line, that v6 is "the fruit of the same poisoned tree."
That's the case in one sentence. Everything else is detail.
Suno's v6 family, which includes the flagship v6 model, the more experimental v6-Wild, and the free v6-mini, launched on September 9, according to Music Business Worldwide, TechCrunch, and Axios. Warner Music Group struck the first peace deal in November 2025, settling its part of the original 2024 lawsuit. BMG followed. Believe, along with its self-releasing platform TuneCore, signed on September 8, one day before launch.
Suno launched its Spark incubator on June 25 offering indie artists cash grants and marketing support, but buried a permanent non-disparagement clause in the contract. With a major copyright lawsuit heading to a July summary judgment hearing and forensic discovery revealing millions of allegedly stolen recordings in its training data, the program... - artist grants with non-disparagement clause - Suno Spark program legal terms
UMG and Sony didn't take the deal. Music Business Worldwide reported that Suno chief product officer Jack Brody said the new model's data doesn't include Universal or Sony repertoire. That split matters because Suno is now running two tracks at once: a licensed product built with Warner, BMG, and Believe, and a live court fight with the two biggest music companies that stayed outside the arrangement.
This isn't Suno's first time in court. Sony and Universal sued the company in June 2024 alongside Warner, over claims that it trained music models on copyrighted tracks without permission. Warner's settlement pulled it out of that fight. UMG and Sony stayed in. This new filing is their answer to Suno's attempt to move on with a fresh, licensed model.
Here's the number that should worry Suno's investors more than the legal theory itself: Music Business Worldwide reported that the 60,202-recording tally could carry a theoretical maximum of just over $9 billion in statutory damages, based on the U.S. copyright law ceiling of $150,000 per work for willful infringement. The labels are also seeking up to $2,500 for each alleged act of circumventing YouTube's anti-down technology.
That's not a rounding error for a startup. That's an existential number.
The new fight is about model lineage #
The legal theory here is the part worth watching closest, because it isn't really about Suno anymore. If a court accepts that retraining a new model on an old model's outputs doesn't clear the underlying infringement, that logic doesn't stop at music. It doesn't stop at songs. Every AI company that has ever fine-tuned, distilled, bootstrapped, or otherwise rebuilt a newer model using an older, legally shakier one now has a template for what a plaintiff's complaint looks like.
Frankly, that's the real stakes of this filing. Suno has said v6 was trained from scratch on licensed material and user data. Axios reported Brody's explanation that the company also used technical and preference learnings from earlier models, while Music Business Worldwide quoted him saying v6 was trained "entirely from scratch, from the ground up." UMG and Sony are trying to make the lineage matter anyway.
That is a hard argument for any AI company to brush aside. You can license the next dataset and retire the old model. You can announce a new product with better names and better partners. But if a court says the old model's outputs carry the taint forward, a relaunch doesn't work like a legal reset button.
Universal Music Group Sues DistroKid Over AI Slop Flooding Streaming Universal Music Group sued DistroKid on September 15, 2026, in Delaware federal court, accusing the distributor of knowingly flooding streaming platforms with AI-generated music and misrepresenting it as human-made. The suit names 1,000 tracks and seeks damages that could reach $150 million, echoing UMG's earlier fight with TuneCore. - AI generated music flooding streaming services lawsuits - DistroKid sued Universal Music Group copyright infringement claims
Suno has not yet answered the new complaint in court. The company has previously defended its training as fair use, and the older Massachusetts case is still moving toward the central fight over whether training an AI model on copyrighted recordings without a license is lawful. In August, Judge F. Dennis Saylor IV allowed Sony and UMG to add a separate YouTube stream-ripping claim to the earlier case, while refusing to fold more than 61,000 additional recordings into that same lawsuit. The second case is how the labels brought those works back.
There is a business point buried inside the legal one. Suno has already shown that licensing deals can be signed - Warner did it, then BMG, then Believe. Sony and UMG are using that fact against it. If a market now exists for training rights, the labels can argue that Suno didn't need to scrape first and settle later.
This case will take months, maybe years, to resolve. But the question is already clear enough: does a new training pass erase what came before, or does the old model leave fingerprints on the new one? Every major AI lab training on scraped, licensed, user-generated, or model-generated content should be watching the answer.
Also read: A Ukrainian Veteran Built a Mental Wellness App Around a Reset, Not Another Routine • Anthropic Hires Accenture to Sit Inside Its Walls and Hunt for Dangerous AI • Lam Research Raises Its Dividend 27% as Analysts See More Room to Run
This article is posted in AI News, check it out for more related stories.
Join the discussion #
Open in the community → Almost there. Sign in and your reply posts straight away.