Suno has disclosed that its new V6 AI music models were trained partly on “creations” from its own community, adding a new layer to its escalating copyright fight with Universal Music Group and Sony Music Entertainment.
The AI music company said September 22nd that V6 was trained using licensed content from music-industry partners, community interactions including “creations and preference signals,” and accumulated technical work from Suno’s team.
The wording matters because it goes further than what Suno publicly described when V6 launched September 9th. Chief Product Officer Jack Brody said at the time that the models had been trained “entirely from scratch, from the ground up” using licensed material, user data and Suno’s technical advances. Suno subsequently clarified that the user data Brody referenced meant preference information rather than audio uploaded by users.
Its September 22nd statement explicitly places user “creations” alongside those preference signals, Music Business Worldwide notes. Suno has not said in that statement that users’ independently uploaded original recordings were used to train V6, and generated creations should not be treated as synonymous with uploads.
The disclosure comes four days after Universal and Sony filed a second federal lawsuit against Suno alleging infringement involving 60,202 copyrighted sound recordings. More unusually, the labels argue that V6 remains connected to Suno’s earlier models even if its direct training mix includes licensed music.
User “Creations” Enter the V6 Training Story
Suno’s September 22nd description identifies three broad ingredients behind V6.
One is licensed material connected with music-industry partners. Another is what Suno describes as the accumulated research and technical work of its team. The third is community interaction, including creations and preference signals.
Preference signals can provide information about which outputs users favor without being audio recordings themselves. “Creations” introduce a separate category: music produced within Suno’s ecosystem that can potentially become material for improving later systems.
The company has not publicly provided a detailed inventory showing which community creations entered V6 training, how many were used or precisely how they were selected.
That distinction matters because Suno allows users to work with generated music as well as their own inputs. Saying V6 incorporated community creations does not establish that every type of audio a user can bring into Suno was included in the model’s training corpus.
What the new disclosure establishes is narrower but still significant: generated activity inside Suno’s own ecosystem has become part of the story of how its next generation of models was built.
Sony and Universal Challenge the Model’s Ancestry
The labels’ newest argument reaches beyond the familiar dispute over whether copyrighted recordings can be copied for AI training.
Universal, Sony and affiliated labels filed their second Suno lawsuit September 18th in the U.S. District Court for the District of Massachusetts. The federal docket identifies the case as UMG Recordings, Inc. et al. v. Suno, Inc. and confirms that the plaintiffs demanded a jury trial.
The new action asserts 60,202 sound recordings. V6 is central to the labels’ theory even though Suno says the model family was developed from scratch with industry partners.
The labels argue that a model’s ancestry matters.
Variety’s review of the complaint describes Universal and Sony’s position that V6 remains connected to allegedly infringing predecessor models because outputs from those systems entered the development process for the new generation.
Under that theory, starting a new model architecture would not necessarily create a clean legal break if the information used to develop it was produced by earlier models whose own training remains disputed.
Suno rejects that characterization. In its September 22nd response, the company called the claims “fundamentally flawed on both the facts and the law.”
Synthetic Music Creates a New Copyright Question
One part of the dispute concerns synthetic training material: audio generated by AI rather than the original copyrighted recordings that may have been used to develop an earlier system.
Universal and Sony allege that V6 may have been trained using synthetic audio created or modified by previous Suno models, along with preference information associated with those outputs. Those are allegations in the complaint and have not been established by a court.
The labels’ position is that generated recordings cannot necessarily be viewed in isolation if the systems that produced them were themselves trained on copyrighted recordings without permission.
That creates a different question from whether V6 directly ingested a Sony or Universal master.
A newer model can potentially learn from large quantities of machine-generated music rather than returning to the same original recording collection. The dispute asks whether that additional step meaningfully separates the new training process from alleged infringement involving its predecessor.
Suno continues to dispute the labels’ claims and has defended its earlier training practices as fair use.
The issue could extend beyond one AI music company. Synthetic data offers developers a way to train later generations of models partly on material generated by earlier ones, making the legal treatment of that lineage increasingly relevant as AI systems evolve.
Knowledge Distillation Raises a Different Issue
The complaint also presents a model-transfer theory that does not necessarily depend on feeding V6 synthetic songs as training audio.
Universal and Sony allege, on information and belief, that Suno may have used knowledge distillation or related techniques while developing V6.
In simplified terms, distillation allows a newer “student” model to learn behavior from an existing “teacher” model. Instead of relying solely on a collection of generated audio files, knowledge can be transferred through the way one model teaches another to reproduce useful capabilities.
The labels argue that if earlier Suno systems developed those capabilities through unauthorized use of copyrighted recordings, transferring them into V6 would preserve a connection to the disputed training.
Suno has not accepted that theory, and the allegation that V6 used the specific model-transfer methods described by the plaintiffs has not been adjudicated.
The dispute therefore raises two related but distinct questions. Synthetic-data claims focus on what generated material may have entered V6 training. The distillation argument focuses on whether capabilities can move between model generations even without recreating the original training process.
Whether either path carries copyright consequences is now part of the fight.
More Than 60,000 Recordings Are in the Case
The second lawsuit dramatically increases the number of works at issue.
Universal, Sony and their affiliated labels accuse Suno of infringing 60,202 sound recordings. The case followed an attempt to add tens of thousands of works to the original litigation, which already involved hundreds of asserted recordings.
Judge F. Dennis Saylor IV declined in August to expand the existing case on that scale, leaving the labels able to pursue additional claims separately. The September 18th complaint followed in the same Massachusetts federal court, where Saylor is also assigned to the new case.
The labels say they identified the additional recordings after obtaining information about Suno’s training data through discovery and using audio-fingerprinting technology, MBW details.
The number creates an enormous theoretical damages ceiling. U.S. copyright law can allow statutory damages of up to $150,000 per work for willful infringement. Applied mechanically to 60,202 recordings, that would exceed $9 billion.
That figure is not a prediction of what Universal and Sony will recover. Liability has not been established, willfulness would have to be addressed, and courts do not automatically award the statutory maximum for every asserted work.
It does show why expanding the dispute from hundreds of recordings to more than 60,000 materially changes the financial scale of the litigation.
YouTube Creates a Separate Legal Fight
The labels are also challenging how Suno obtained some of the audio used in training.
In a September 1st filing in the original litigation, Suno acknowledged that training audio had been obtained from YouTube and identified YT-DLP in responding to allegations about the downloading process. That factual acknowledgment does not resolve whether the method violated copyright law.
Universal and Sony allege that Suno circumvented technological protections intended to prevent direct downloading of YouTube media. The new complaint therefore includes a legal theory separate from the underlying claim that copying recordings for AI training infringed copyright.
Bloomberg Law identifies the alleged stream-ripping activity as a separate part of the September 18th action covering the 60,202 recordings.
Suno disputes the labels’ legal position and has challenged whether they can pursue the YouTube-related claims. Its broader fair-use defense over AI training remains contested as well.
The plaintiffs are seeking damages and an injunction and have demanded a jury trial. No merits ruling has been issued in the new case.
Some Rights Holders Sue While Others License
The dispute is particularly unusual because Suno is no longer operating entirely outside the licensed record business.
Warner Music Group, originally a plaintiff alongside Universal and Sony in the earlier litigation, settled with Suno in 2025 and entered a licensing relationship with the company. Suno later announced partnerships with BMG and Believe.
When V6 launched September 9th, Suno described it as developed with industry partners including Warner Music Group, BMG and Believe. The company positioned those relationships as a blueprint for AI developers and the music business to build products together.
Universal and Sony see those licensing relationships differently. Their complaint points to deals between Suno and other rights holders as evidence that a commercial market exists for obtaining permission around AI music development.
Suno disputes the labels’ characterization of those agreements. Brody said around the V6 launch that it would be misleading to view the partnerships simply as transactions for training data, arguing that they extend into broader product and commercial collaboration.
The split leaves Suno in an extraordinary position. The same company is developing its newest AI music systems alongside licensed music-industry partners while defending federal lawsuits brought by other major rightsholders over the technology that preceded them.
September 22nd adds user creations to that picture.
Suno says V6 was built from scratch using licensed material, community creations and preference signals, plus its team’s accumulated technical work. Universal and Sony argue that a technically new model cannot necessarily escape the disputed history of the systems and outputs that helped lead to it.
The question is no longer only what an AI music model trained on.
It is whether a new model can ever completely leave its training ancestry behind.