Sony v. Suno — Copyright | Outside IP Counsel

Major labels accuse the AI music generators Suno and Udio of copying sound recordings to train their models. The pending Boston and New York cases test fair use for AI audio.

In June 2024, the recording industry opened a second front in the AI copyright wars. The major labels — led by Universal Music Group and Sony Music, joined by Warner — sued the two best-known AI music generators in twin actions: UMG Recordings, Inc. v. Suno, Inc., No. 1:24-cv-11611 (D. Mass., filed June 24, 2024), and UMG Recordings, Inc. v. Uncharted Labs, Inc. (d/b/a Udio), No. 1:24-cv-04777 (S.D.N.Y., filed June 24, 2024). The complaints, coordinated with the Recording Industry Association of America, allege that both companies copied enormous quantities of copyrighted sound recordings to train models that now spit out music imitating the real thing. For the songwriters, session musicians, and producers who make Los Angeles a music capital, these cases will help decide whether generative audio is built on licensed foundations or on uncompensated copying. This matter remains unsettled as of June 29, 2026, and nothing here should be read as a final ruling.

At a glance

Cases: UMG Recordings, Inc., et al. v. Suno, Inc., No. 1:24-cv-11611 (D. Mass.), before Judge F. Dennis Saylor IV; and UMG Recordings, Inc., et al. v. Uncharted Labs, Inc. (d/b/a Udio), No. 1:24-cv-04777 (S.D.N.Y.), before Judge Alvin K. Hellerstein.
Plaintiffs: UMG Recordings, Sony Music Entertainment, Warner Records, and affiliated labels, coordinated by the RIAA.
Posture: Pleadings and discovery; amended complaints filed in 2025 added stream-ripping and DMCA allegations.
Status: Pending — Universal settled with Udio (Oct. 2025) and Warner settled with Suno (Nov. 2025), but Sony and others continue to litigate; a Suno summary-judgment hearing was set for July 2026 — unsettled as of June 29, 2026.
Significance: The first major label test of whether copying sound recordings to train generative-music models is infringement or fair use.

The core allegation: copying the catalog to train the model

Both complaints rest on a direct-infringement theory: that Suno and Udio reproduced the labels’ copyrighted sound recordings — potentially in the millions — to train their models, and that this copying is itself infringement of the reproduction right under Section 106 of the Copyright Act, separate from any question about outputs. To illustrate, the complaints point to model outputs that allegedly reproduce recognizable elements of famous recordings, from vocal timbres to distinctive musical passages, as circumstantial evidence that protected recordings were ingested.

The labels seek statutory damages of up to $150,000 per infringed work under Section 504(c), a figure that, multiplied across thousands of recordings, becomes existential for the defendants. They also seek injunctive relief. The defendants, for their part, have not denied training on copyrighted material so much as defended it: both have argued that ingesting recordings to learn musical patterns is a transformative fair use, akin to a person learning to play by listening.

The fair-use battleground

Fair use is the decisive question, and it is unresolved. The companies analogize training to human learning and to the kind of intermediate copying courts have sometimes blessed in software cases. The labels counter that the models compete directly in the market for music — the fourth fair-use factor — and that wholesale copying of entire recordings, for a commercial product that generates substitute songs, is the opposite of transformative.

The audio context sharpens the fourth-factor argument. Where text-training cases can debate whether a chatbot competes with the books it read, an AI that generates songs plainly occupies the same market as the recordings it learned from. That market-substitution concern is why the labels frame these as cases about the survival of a licensing market for music, not abstract questions about machine learning. A summary-judgment hearing in the Suno matter was scheduled for July 2026, putting the fair-use question on track for a substantive ruling.

Stream-ripping, DMCA claims, and the settlements

As discovery progressed, the labels sharpened their factual theory. Amended complaints in 2025 alleged that the defendants obtained training audio in part by “stream-ripping” recordings from YouTube — extracting audio in violation of platform protections — and added claims under the DMCA’s anti-circumvention provisions. Those allegations, if proven, would undercut a fair-use defense by casting the acquisition of training data as itself unlawful.

The litigation has also begun to fracture into settlements. Universal Music Group settled with Udio in October 2025, pairing the resolution with a licensing arrangement for a future AI music product; Warner Music settled with Suno in November 2025. But other plaintiffs, notably Sony Music, declined to settle, evidently preferring a court precedent on liability to a private deal. The result is a hybrid landscape: licensed cooperation on one track, contested litigation on another, with no controlling merits ruling yet.

Open questions

Is training on sound recordings fair use? No court has decided it for AI audio; the July 2026 Suno hearing may produce the first substantive answer.
How much does data provenance matter? If the defendants stream-ripped recordings, that unlawful acquisition could weigh heavily against fair use and add independent DMCA liability.
Will settlements set the market price? Per-generation royalty frameworks emerging from the settlements may shape licensing norms even before any court rules.

Implications

For labels and artists: Direct copying of recordings for training is the strongest claim, and market substitution is the strongest factor — but both must survive a fair-use defense that has not yet been adjudicated.
For AI music developers: How training data was obtained is now central. Lawful licensing or clean sourcing materially strengthens a fair-use position; stream-ripping allegations gravely weaken it.
For licensing: The Udio and Suno settlements suggest a licensed-platform model is forming in parallel with the litigation, which may matter more commercially than any single ruling.
For litigators: Plead acquisition-side conduct (circumvention, stream-ripping) alongside the core reproduction claim; it can blunt fair use and open additional statutory remedies.
For everyone watching: These are active cases with mixed settlements. Nothing here is a final ruling, and the unsettled claims could come out either way.

Frequently asked questions

Have the courts decided whether training AI on sound recordings is legal? No. Both cases were filed in June 2024 and remain in litigation. The companies deny infringement and assert fair use, and no court has issued a final merits ruling. A summary-judgment hearing in the Suno case was set for July 2026. The matter is unsettled.

Did some of these cases settle? Yes, in part. Universal Music Group reached a settlement and licensing arrangement with Udio in October 2025, and Warner Music settled with Suno in November 2025. But Sony Music and other plaintiffs continued to litigate, so live claims against both companies remain.

What exactly do the labels accuse Suno and Udio of copying? The complaints allege the companies copied vast numbers of the labels’ copyrighted sound recordings to train their models, producing outputs that imitate real recordings. Later amended complaints added allegations of stream-ripping from YouTube and related circumvention claims.