The $9 Billion Sequel: Universal and Sony Sue Suno Again, Calling v6 'the Fruit of the Same Poisoned Tree'

On September 18, 2026, Universal Music Group and Sony Music Entertainment walked into the federal courthouse in Boston and sued Suno for the second time. The new complaint, filed as case 1:26-cv-14275 in the District of Massachusetts, asserts 60,202 sound recordings that the labels say Suno copied without a license and used to build the models behind its AI music generation service. At the statutory maximum of $150,000 per willfully infringed work, the ceiling on this single filing runs past $9 billion — and the labels are explicit that the 60,202 count is “only a small portion” of what they believe is out there.

The filing lands nine days after Suno launched v6, its most defensible model ever: a suite trained “entirely from scratch” on licensed catalogs from Warner Music Group, BMG, and Believe, with revenue sharing turned on from day one. That launch was supposed to be Suno’s clean slate. Universal and Sony’s answer is the most interesting legal argument the AI music wars have produced so far: it does not matter that v6’s training data is licensed, because v6 was built on the outputs of earlier Suno models that were not. In the words of the complaint, v6 is “not a fresh start; it is the fruit of the same poisoned tree.”

Why there is a second case at all

The first UMG/Sony suit against Suno, filed in June 2024, is still running and covers 560 works. During discovery, the labels commissioned audio-fingerprinting company Audible Magic to scan Suno’s actual training corpus, and the analysis turned up more than 61,000 additional recordings belonging to them. They moved to amend the original complaint to add all of them.

Judge F. Dennis Saylor IV denied that motion on August 18, 2026 — not because he doubted the claims, but because bolting 61,000 works onto a case he wants decided on fair use would blow up the schedule. The labels, he wrote, are “of course entitled to pursue valid claims for copyright infringement,” and pointed them toward parallel proceedings. The new complaint frames itself as doing exactly what the court contemplated: a second case carrying the forensic findings, while the first case races toward its fair-use ruling with fact discovery closing September 30.

The laundering theory

The core allegation is about lineage. Suno’s Chief Product Officer Jack Brody said at the v6 launch that the new models were trained from the ground up on data that “doesn’t include data from Universal or Sony.” The labels do not dispute that v6’s direct training corpus was clean. They allege instead that v6 was distilled from models that were not.

The complaint describes two transmission mechanisms. The first is “user interactions”: Suno has admitted that v6 was trained partly on how users responded to previous models. But Suno generates two tracks per prompt and records which one the user prefers — and every one of those preference signals, the labels argue, is “a judgment about audio outputs generated using Suno’s unlicensed models,” synthetic outputs that “embody the expressive features that Suno’s tainted models derived from Plaintiffs’ recordings.”

The second is knowledge distillation proper, which the filing describes as a process “whereby a new ‘student’ model is trained to reproduce the learned behavior of a predecessor ‘teacher’ model.” The teachers here are Suno’s earlier releases — the complaint names v4, v4.5, v4.5+, v5, v4.5-all, and v5.5 — and the capabilities they transmit, “how to compose, arrange, and render convincing imitations of human-created music across genres and styles,” were derived from unlicensed recordings. The filing adds that Suno has never said it destroyed the copied recordings, so “every model Suno has ever released — from its earliest versions through v6 — is the product of, and continues to exploit, Suno’s unauthorized copying.”

This is the argument with reach far beyond music. Distillation from a predecessor model is standard practice across the entire AI industry — it is how most labs build cheaper, faster successors to their flagships. If a court accepts that training a new model on an old model’s outputs “launders” the old model’s infringement, the theory stops being about one music startup and becomes a question every AI company with a messy training-data history has to answer. A CISA advisory two weeks ago alleged Chinese labs used the same technique against US models; Cognition built SWE-2 on a distilled model. The doctrine, if it takes hold, will be cited in text-model cases before the year is out.

Suno’s own deals and terms, weaponized

The sharpest sections of the complaint are built from Suno’s own conduct. The filing lists three licensing agreements signed in under a year — Warner in November 2025, BMG on August 12, 2026, and Believe on September 8 — and argues they “confirm that a functioning market exists for licensing sound recordings to train generative AI models and that Suno itself recognizes that using copyrighted sound recordings for this purpose requires permission.” Fair use analysis weighs market harm, and a defendant who simultaneously pays three rightsholders for training data has a harder time arguing no such market exists. Suno has tried to frame the deals differently, with Brody saying the revenue share is “not in exchange for training” and “not really about the data” — a characterization the complaint calls “self-serving.”

The labels also turn Suno’s user contract against it. Since January 2024, Suno’s terms of service have demanded from every user a “worldwide, non-exclusive, fully paid-up, sublicensable” and “irrevocable” license to reproduce, store, modify, and create derivative works from uploaded content. Demanding a license upstream while taking commercial recordings for free, the argument runs, shows Suno knew training on someone’s audio requires permission — which is what makes the infringement “knowing and willful,” the finding that unlocks the $150,000-per-work tier.

Even the artist-name filter gets a mention. Suno once blocked prompts containing specific artists’ names and cited the block as proof its models generated “completely new outputs.” The complaint says the filter is gone: “This design feature turned out to be so fundamental that Suno simply switched it off,” with artist names now usable through prompt augmentation.

The numbers behind the market-harm claim

On harm, the filing reaches for measurement rather than adjectives. Deezer reported in July 2026 that AI-generated tracks passed half of all daily uploads for the first time — roughly 90,000 per day, up from about 39 percent six months earlier. Suno has told investors its users generate a Spotify catalogue’s worth of output every two weeks. “Every machine-generated track delivered to a streaming service competes with the Copyrighted Recordings for placement, for discovery, for finite listener attention, and for a share of the fixed royalty pools from which genuine artists and rightsholders are paid,” the complaint states. “Dilution at this scale and velocity is not just a future harm; it is a present, measured, and accelerating one.”

The financial stakes are laid out too. Suno’s latest round raised more than $400 million at a $5.4 billion valuation, up from $2.45 billion in autumn 2025. The labels say Suno makes hundreds of millions a year and projects $1 billion in revenue by 2028.

Three counts, and what comes next

The complaint pleads two counts of direct infringement — split between post-1972 recordings and pre-1972 works protected under the Music Modernization Act — and a third for circumvention, alleging Suno “stream ripped” audio from YouTube using YT-DLP in breach of YouTube’s terms and Section 1201(a), a claim carrying up to $2,500 per act (roughly $150 million applied across the asserted recordings). Alongside damages, the labels seek an injunction and a jury trial. Suno has not yet responded, and its defense in the first case remains that training on copyrighted recordings is “quintessential fair use.”

Nobody expects a $9 billion check. The number is designed to force a license, and the pattern around it is unmistakable: UMG sued distributor DistroKid on September 15, settled into a licensed deal with ElevenLabs on September 10, and has now sued the generator that declined to sign. Warner, BMG, and Believe took the deal; Universal and Sony are litigating for better terms. The most likely endgame is a settlement that looks like the ElevenLabs arrangement. But the “poisoned tree” theory is the part worth watching regardless of how Suno resolves it — because if distillation laundering becomes doctrine, it will not stay confined to music.