Suno has admitted that it has not deleted all of the sound recordings it used to train its AI models.
The admission appears in Suno‘s answer to the labels’ second copyright lawsuit, filed on September 18, which alleges that, by training its new v6 models on the outputs of earlier ones, Suno “launders” its infringement of their recordings.
The 44-page answer was filed on Tuesday (October 20) in the US District Court for the District of Massachusetts, and reviewed by MBW.
You can read Suno‘s answer in full here.
The labels’ complaint alleges that Suno “has never claimed to have destroyed or ceased using” the copies of copyrighted works it obtained, and that “on information and belief it retains and continues to use them.”
Suno‘s answer admits that it “has not deleted all of the sound recordings it has used to train its AI models,” and says this is “consistent with Plaintiffs’ demand in correspondence prior to the Original Action that Suno retain ‘data sets used to train’ Suno’s AI models.”
The answer does not identify that correspondence. In their complaint, the labels say they wrote to Suno on April 25, 2024, asking it to “cease and desist from copying or otherwise using their copyrighted sound recordings” and to provide a complete list of the materials it used to develop its AI models.
What the answer does not say is that Suno has stopped using the recordings, or that it is keeping them only because of the litigation. Suno disputes the claim that it “continues to use them” only in general terms, closing its response by denying “any remaining allegations of this paragraph.”
That matters to the labels’ case against v6. Their complaint alleges, “on information and belief,” that Suno‘s development of the new models “drew on these retained copies.”
“Training a ‘new’ model on the outputs of an infringing model does not eliminate the infringement; it launders it.”
UMG and Sony complaint against Suno
The retention admission appears six times in the answer, including in Suno‘s responses to both of the labels’ copyright infringement claims.
The “Original Action” is the labels’ first case, filed in June 2024, in which Suno said in an August 2024 answer that the relevant data had all “been preserved.”
In the new answer, Suno admits that the labels inspected its source code and training data during discovery in that case.
It also admits that their analysis of the data using Audible Magic “purported to identify millions of copyrighted works Plaintiffs claim to own or control in Suno’s training dataset, including the Copyrighted Recordings,” the 60,202 recordings asserted in the new suit.
Suno says it lacks knowledge about the labels’ process and their ownership of the recordings, and denies those allegations on that basis.
Suno also admits that users have been able to include artist names in their prompts since January 2026, after it previously prevented them from doing so.
Suno admits that a prompt “can include specifying the artist (which is converted into a related style), genre, lyrics, story direction, and themes to serve as inspiration,” and denies that such prompts are used to generate outputs “without first converting the artist name to a style description.”
In its August 2024 answer, Suno told the same court that its tool “has been designed not to be able to process requests for new outputs when they are described by reference to a particular artist’s name, rather than by reference to a musical style, in general.”
The labels’ complaint says that after the first suit was filed, Suno CEO Mikey Shulman pointed to the ban as proof that Suno‘s models were “designed to generate completely new outputs.” That design feature, it adds, “turned out to be so fundamental that Suno simply switched it off and now permits users to freely prompt its service using names of specific recording artists through prompt augmentation.”
Suno admits lifting the ban, but “denies that it ‘switched off’ its model design to ‘generate completely new outputs.'”
Suno also denies the complaint’s claim that the earlier ban was “a fig leaf to conceal what the models had learned.”
“When someone asks for a particular artist, we translate that into the musical qualities behind the request – not so that a user can imitate someone else’s work, but so that people learn the language of music and make something of their own.”
Suno, responding to the labels’ second lawsuit
In a blog post on August 6 this year, Shulman wrote that Suno has “never allowed prompts for specific artists or copyrighted songs.” He added: “Instead, when a prompt references a specific artist, we remove the artist’s name and redirect the request toward descriptive musical characteristics.”
Responding to the second suit on September 18, Suno said: “When someone asks for a particular artist, we translate that into the musical qualities behind the request – not so that a user can imitate someone else’s work, but so that people learn the language of music and make something of their own.”
Artist-name prompts are also central to a separate class action brought by Jason Isbell and other musicians, which defines most of its proposed subclasses by musicians whose names “are or have been accepted as prompts in Suno.” Suno‘s response in that case is due on November 3.
The labels’ central new allegation concerns v6, the model suite Suno launched on September 9 with Warner Music Group, BMG, and Believe.
Suno has said that v6 “was trained on content licensed from our partners, interactions including creations and preference signals from our community, and the accumulated learnings from our team,” a statement it repeats several times in its answer.
The labels allege, “on information and belief,” that Suno “trained v6 on synthetic audio files that users generated or altered with its prior tainted models, with preference data that is keyed to those outputs, or both.”
“Training a ‘new’ model on the outputs of an infringing model does not eliminate the infringement; it launders it,” the Universal and Sony complaint states. “…v6 is not a fresh start; it is the fruit of the same poisoned tree.”
Suno “denies that each of its models was trained on the same corpus of sound recordings,” as the complaint alleges of its models from v4 to v5.5.
Suno admits that it “uses information about user responses to its service to improve its service,” and that it “maintains certain data concerning user-generated outputs and their preference signals.”
The answer does not respond by name to the complaint’s knowledge distillation allegations, which appear in the same paragraph as the “retained copies” claim. Its response to that paragraph says that, to the extent a response is required, Suno “lacks knowledge or information sufficient to form a belief” about the allegations, “and on that basis denies them.”
Elsewhere, Suno “denies that its model is designed to generate outputs that replicate, reproduce, or sound like existing sound recordings,” and argues that any copying is fair use.
When the second suit was filed, Suno said the labels’ claims “remain fundamentally flawed on both the facts and the law.”
Suno answered the complaint rather than moving to dismiss it. Its answer asks the court to enter judgment in its favor, dismiss all of the labels’ claims with prejudice, and award it attorneys’ fees and costs.
MBW has reached out to Universal Music Group and Sony Music Entertainment for comment.Music Business Worldwide
