Sony and UMG sue Suno again, saying licensed v6 inherited old infringement

Why it matters

The case could decide whether a licensed AI model remains exposed when its feedback data and technical lineage come from disputed predecessors.

Sony Music Entertainment and Universal Music Group sued Suno on Friday, alleging that its newly licensed v6 music models still benefit from copyrighted recordings used without permission in earlier versions. The case attacks the central bet CEO and co-founder Mikey Shulman has made this year: that Suno can bring record labels inside its model-development process without abandoning the technology or mass-market product that made it valuable.

The complaint filed on September 18th in Massachusetts federal court identifies 60,202 sound recordings that the labels claim Suno copied. It arrived nine days after Shulman introduced v6, v6-wild and v6-mini as a new model generation developed with Warner Music Group, BMG and Believe.

Suno told Engadget that the new claims are "fundamentally flawed on both the facts and the law." Suno said v6 was trained using licensed content from its partners, creations and preference signals from users, and technical learnings accumulated by Suno's researchers.

That description supplies the labels with their new line of attack.

The labels are attacking Suno's reset

Sony and UMG allege that Suno carried disputed material into v6 through user interactions with older models. When Suno generated two songs for a prompt, a user's selection between them became a preference signal that could help improve later models. The labels argue that those signals remain tied to outputs produced by systems trained on their recordings.

The complaint also alleges that Suno used knowledge distillation, in which a new model learns behavior from an earlier model. Under the labels' theory, building v6 from licensed recordings would not cleanly separate it from earlier systems if those systems supplied outputs, rankings or learned capabilities during development.

Those are allegations, and the court has not ruled that model outputs, preference data or distilled technical knowledge carry copyright liability from one generation to the next. The case pushes directly into that unresolved territory. A ruling for the labels could make replacing a disputed training set far harder than deleting files and starting a new training run.

Suno presented v6's September 9th launch as a product and business reset. The flagship v6 and experimental v6-wild models are available to paying subscribers, while the faster v6-mini is open to all users. The models support section-level editing, sampling, mashups and creation from text, audio, images and video. Suno also said it would retire its previous models.

Shulman's announcement described v6 as a foundation for opt-in artist products that would pay participating musicians. The lawsuit challenges whether that forward-looking licensing structure can be separated from how Suno developed its earlier technology.

Suno's licensing push becomes part of the labels' case

Shulman founded Suno with Georg Kucsko, Martin Camacho and Keenan Freyberg after the four worked together at Kensho Technologies, the Cambridge AI developer acquired by S&P Global. Shulman and Camacho also played music together, and their experiments applying audio-model expertise to songs became the basis for Suno.

The founders built Suno around a simple consumer thesis: people should be able to make complete songs without learning an instrument or production suite. As Axios reported when Suno emerged from stealth, the product generated two songs from a written description and was designed for users with little formal musical training.

That accessibility helped Suno move from a prompt-based novelty toward a fuller creation suite with editing, audio uploads, stem exports and a browser-based workstation. It also produced the scale that made licensing commercially possible and legally urgent.

Sony, UMG and Warner first sued Suno in June 2024. Suno argued that training on copyrighted recordings qualified as fair use. Warner later left that fight after reaching a licensing agreement with Suno in November 2025. BMG partnered with Suno in August 2026, followed by Believe on September 8th.

Sony and UMG now cite those deals as evidence that a functioning market exists for licensing recordings used in AI training. Their argument puts Shulman's transition in a bind: each agreement strengthens Suno's ability to work with artists and rights holders, while also giving the remaining plaintiffs a market price and precedent for the permissions they say Suno should have obtained earlier.

The labels are also using Suno's development language to support their allegations. References to accumulated learnings and community preferences sound routine in machine-learning product work. In this case, the plaintiffs cite them to argue that a model can inherit legal exposure without receiving the original audio files in its final training corpus.

A second lawsuit, with a larger catalog

Friday's case is the second copyright action filed against Suno by Sony and UMG. The first case covers 560 works. The labels sought to add more than 61,000 recordings, but US District Judge F. Dennis Saylor IV denied that request without prejudice on August 18th because it would disrupt the existing schedule. The judge left the labels free to pursue those claims separately, leading to the new complaint.

The procedural history matters because Friday's filing does not represent a second judicial finding against Suno. It gives the labels another vehicle for litigating a much larger catalog while the original fair-use case continues.

The new complaint requests statutory damages of up to $150,000 for each infringed work. Applied to all 60,202 recordings, the theoretical ceiling exceeds $9 billion. Sony and UMG also seek up to $2,500 for each alleged act of circumventing YouTube's download restrictions and an injunction against further use of their recordings.

The anti-circumvention allegations draw support from source code and internal material obtained in a July hack. 404 Media reported that the material described scraping from YouTube Music, Deezer, Genius and other sources. Suno said at the time that the exposed code was outdated and no longer used. RuntimeWire previously detailed what the hacked code showed about Suno's data collection.

Suno has since added fingerprinting, watermarking and download restrictions intended to curb unauthorized distribution and improve provenance, as RuntimeWire reported in August. Those safeguards address how generated tracks move through streaming services. The latest complaint reaches further back into the lineage of the models themselves.

A $5.4 billion transition

The stakes extend beyond legal damages. In June, Suno said it had raised more than $400 million at a $5.4 billion post-money valuation. Bond Capital led the Series D, joined by IVP, Forerunner, Union Square Ventures, Alkeon and Quiet. Matrix, Lightspeed, Menlo Ventures and Schroders Capital also participated.

As RuntimeWire reported when the round closed, Shulman was already pushing Suno beyond one-click song generation and toward tools for artists, producers and fans. Licensing became part of that expansion rather than a retreat from it.

Sony and UMG allege that Suno's technological continuity means the reset never happened. Suno's defense depends on establishing a meaningful boundary between disputed earlier training and v6's licensed development process.

Shulman has moved Suno closer to the music industry while preserving the founders' original goal of giving anyone a way to make songs. The lawsuit will test whether that transition came early enough, and whether an AI model can ever leave its predecessors' training history behind.