Suno Has Not Deleted All Training Recordings

Suno has admitted in a court filing that it has not deleted all of the sound recordings it used to train its AI music models, as Universal Music and Sony press a second lawsuit alleging that its newest models “launder” earlier infringement.

The admission is in Suno’s 44-page answer, filed on Tuesday 20 October in the US District Court for the District of Massachusetts. Suno says it “has not deleted all of the sound recordings it has used to train its AI models, consistent with Plaintiffs’ demand in correspondence prior to the Original Action that Suno retain ‘data sets used to train’ Suno’s AI models.” The same wording recurs throughout the answer, including in its responses to both infringement claims. Music Business Worldwide first reported the filing.

The labels say v6 is ‘the fruit of the same poisoned tree’

The labels’ complaint, filed on 18 September by UMG Recordings, Capitol Records, Sony Music Entertainment and a group of Sony labels, describes itself as their second action against Suno. It asserts 60,202 recordings, which it calls “only a small portion” of those Suno copied, and alleges the company obtained audio by ripping it from YouTube in circumvention of technical protections.

Its central new claim concerns v6, the model suite the complaint says Suno launched on 9 September. “Training a ‘new’ model on the outputs of an infringing model does not eliminate the infringement; it launders it,” the complaint states. It adds that Suno “has never claimed to have destroyed or ceased using the unauthorized copies of copyrighted works it unlawfully obtained in the first instance, and on information and belief it retains and continues to use them. v6 is not a fresh start; it is the fruit of the same poisoned tree.”

The labels bring three claims: direct infringement of recordings made after 1972, direct infringement of earlier recordings, and circumvention of technological measures. They ask for a declaration of wilful infringement, an injunction, and statutory damages of up to $150,000 per work infringed and up to $2,500 per act of circumvention.

The complaint says the labels had tried to add more than 61,000 works to their first case, filed in June 2024. The court refused on 18 August, mainly because of the effect on the schedule, while saying the labels “are of course entitled to pursue valid claims for copyright infringement, and the magnitude of the alleged infringement is not a defense.”

Suno admits keeping data but not that it still uses it

The answer does not say Suno has stopped using the retained recordings. According to MBW, Suno disputes the claim that it “continues to use them” only through a general denial of the paragraph’s remaining allegations. MBW also reported that Suno told the court in an August 2024 answer in the first case that the relevant data had all “been preserved”.

Suno repeats its public account of v6, saying the model “was trained on content licensed from our partners, interactions including creations and preference signals from our community, and the accumulated learnings from our team.” It admits that it “obtained audio data from YouTube for use as training data”, and that the labels’ analysis of its training data using Audible Magic “purported to identify millions of copyrighted works Plaintiffs claim to own or control in Suno’s training dataset”. It says it lacks knowledge of that process and denies the allegations on that basis.

The filing also confirms a change in how Suno handles artist names. Suno admits that since January 2026 users have been able to include them in prompts, after it previously blocked this. It denies that such prompts generate output without first converting the name into a style description, and denies that it “switched off” its model design.

Suno’s defence rests on fair use and copyright misuse

Suno answered the complaint rather than moving to dismiss it. Its first defence is fair use: it says its service “uses a back-end technological process, invisible to the public, in the service of creating an ultimately non-infringing new product. This is quintessential fair use.” Its second is copyright misuse and unclean hands, alleging that the labels “have engaged in anticompetitive activities that extend an unlawful monopoly over the production and commercialization of music”. A third cites waiver, estoppel and laches. It asks the court to dismiss the claims with prejudice and award it fees and costs.

The retention admission matters because the complaint’s laundering theory depends on a chain from the original copies to v6. Suno has now confirmed the first link exists in some form, while framing it as compliance with the labels’ own preservation demand. The labels’ complaint, for its part, alleges “on information and belief” that v6 drew on those retained copies; that is an allegation, and Suno’s answer does not concede it. The admission also sits awkwardly with the commercial story Suno has told. The complaint cites Suno’s November 2025 licensing deal with Warner Music Group, a former plaintiff, and its stated plan to “deprecate” its “current models”. A clean break from earlier models is harder to argue if the recordings behind them remain on file. Suno’s answer instead presents retention as what the labels themselves asked for before the first case, when they demanded it keep the “data sets used to train” its models. Whether retention becomes evidence of continued use will turn on discovery into how v6 was actually trained.

The case sits alongside a growing set of music claims against AI companies tracked on TSN’s AI copyright lawsuits hub. On Wednesday, Reservoir sued Anthropic and Suno in California over 1,050 compositions it says were used to train Suno’s song generator. The same day, a federal judge heard Anthropic’s fair-use arguments over Claude and song lyrics without ruling.

MBW reported that Suno’s response in a separate class action brought by Jason Isbell and other musicians is due on 3 November. MBW said it had asked Universal and Sony for comment.