Reservoir Media Management, Inc. v. Suno, Inc.

Who is suing

The plaintiffs are Reservoir Media Management, Inc., a Delaware company based in New York, and two English companies it owns through a UK subsidiary, one of them Shapiro, Bernstein & Co. Limited (¶¶ 15-17). The complaint calls them “Reservoir” and describes Reservoir as an independent music company whose parent trades on NASDAQ (¶¶ 1, 18).

Reservoir says it owns or exclusively controls the U.S. copyrights in 1,050 registered compositions listed in Exhibit A (¶¶ 2, 19). The complaint names examples including Joni Mitchell’s “Big Yellow Taxi” and “Both Sides Now,” Johnny Cash’s “Ring of Fire,” Madonna’s “Papa Don’t Preach,” Miles Davis’s “So What” and four Evanescence songs (¶ 2). It says Reservoir licenses lyrics to aggregators such as LyricFind and Musixmatch on terms that require the title and songwriter credits to be shown with them (¶ 30).

Suno is a Delaware company based in Cambridge, Massachusetts. The complaint describes it as a subscription service that generates complete songs, including vocals and lyrics, from text prompts (¶¶ 3, 5-6, 20, 31). It pleads venue in this district on Suno’s San Francisco office and its use of Oracle servers that it says are located in San Jose (¶¶ 22-23).

How the complaint says Suno got the music

The starting point is Suno’s own words. The complaint quotes Suno’s answer in UMG Recordings, Inc. v. Suno, Inc., No. 1:24-cv-11611 (D. Mass.), stating that its “training data includes essentially all music files of reasonable quality that are accessible on the open Internet, abiding by paywalls, password protections, and the like” (¶ 32). Reservoir argues that, given that scope and how widely its songs are streamed, it is a “near certainty” its Works were copied (¶ 33).

• Stream ripping. The complaint alleges, on information and belief, that Suno used the tools yt-dl and yt-dlp to get around YouTube’s “rolling cipher,” which it describes as an encryption of the underlying file URL, and to download millions of songs (¶¶ 35-38). It says this became public in September 2025, when the ICMP shared private datasets with Billboard (¶ 39).
• Bright Data. Suno is said to have used Bright Data’s proxy networks, “Web Unlocker” and “Scraping Browser” to get past logins, paywalls, CAPTCHAs, rate limits and robots.txt on licensed streaming and lyrics sites (¶¶ 13, 40-42). Bright Data is not a defendant.
• Leaked code. Relying on a 404 Media report, the complaint says leaked Suno source code shows scraping of YouTube Music, Deezer and Genius. It cites a YouTube Music workflow with 2,057,486 artist records, a Genius workflow with 6,980,653 song-page records, and a “youtube_music” file of over two million clips totalling about 113,879 hours (¶¶ 13-14).
• User uploads. The complaint says Suno’s terms give it a “perpetual, irrevocable” license to train on user uploads, and that it therefore trains on copyrighted material users submit (¶¶ 48, 51).

Only a few Works are tied to that code by name. The complaint says “Both Sides Now,” “Diamonds on the Inside,” “Ring of Fire” and “Jukin’ City” appear in the leaked code. It says a separate file holds song metadata for other Works, including “8th of November,” “Good Friends,” a cover of “River,” two Kings of Leon songs and a cover of “Bring Me to Life” (¶ 75). For the rest, it alleges on information and belief that all the Works “exist in some form or fashion” in Suno’s code and training sets (¶ 75). It also says The Atlantic’s AI Watchdog tool shows many of the Works in datasets known to be used for AI training, without naming which ones (¶ 3).

The complaint describes two training techniques from the leaked code. One is “stem splitting,” which separates a recording into vocals, drums and other instruments. Suno is also said to have sought out a cappella versions to model vocals and lyrics more faithfully (¶ 79). The other pairs a cover recording with the original recording of the same song in a single training example, so the model learns to predict the original from the cover (¶ 80).

Three counts, five acts of copying

On CMI, the complaint lists the song title, songwriters, publisher, ISRC codes and copyright notices as they appear on licensed lyrics and streaming platforms (¶ 90). It argues that excluding page headers and footers, where it says CMI “regularly exists,” is the same as removing CMI (¶¶ 40-41). For the knowledge element it says Suno knew its models memorize training data, so stripping CMI would conceal infringement. It cites the October 2025 ruling in Concord v. Anthropic that similar allegations were enough at the pleading stage (¶ 92).

The output allegations

The complaint says Suno’s outputs include verbatim, strikingly similar or substantially similar copies of the Works (¶¶ 52, 78). Most of the examples it gives involve other rightsholders’ songs and come from other proceedings:

• The 2024 UMG complaint’s allegations of outputs resembling “Johnny B. Goode” (27 times), “Rock Around the Clock” (10), “Great Balls of Fire” (10), “The Thrill Is Gone” (6) and “Sway” (11), including identical pitches and rhythm in some phrases (¶¶ 56-60).
• Round Hill Music LP v. Suno, Inc., No. 3:26-cv-8507-RFL (N.D. Cal.), on a near-verbatim output of the lyrics to “I Got You (I Feel Good),” and Gerencia 360 Music, Inc. v. Suno, Inc., No. 3:26-cv-09248-RFL (N.D. Cal.), on an output based on translated lyrics of “Quince Besos” (¶ 50).
• The GEMA case in Germany, where the complaint says GEMA’s lawyers reproduced “Daddy Cool” and “Mambo No. 5” almost note for note by prompting Suno with the lyrics, a style and a title (¶ 55).

The complaint’s own prompt example names an artist rather than a Work. Asked for “a Miles Davis like Blues track with horns and a quartet,” Suno is said to have produced two tracks the complaint calls unlicensed derivatives of “Miles Davis music” (¶ 49). It also alleges that guardrails are easy to get around: a prompt naming Reba McEntire is refused but a misspelling is accepted, third-party services sell artist-specific prompts, and version 6 now redirects artist names to descriptive characteristics instead of refusing them (¶¶ 46-47). The complaint says discovery will show outputs infringing all the Works (¶ 62).

Settlements and rulings the complaint relies on

The complaint says Suno settled with Warner Music Group, agreed to deprecate its then-current models and launch V6, and reached agreements with BMG and Believe to use their music as training data (¶ 10). Reservoir argues these deals show that a market for licensing training data exists (¶ 10). It also lists AI companies and collecting societies that it says pay for training data (¶ 70).

Outside the United States, the complaint says the Munich Regional Court ruled on 31 July 2026 in GEMA’s suit that Suno infringed by training on and reproducing GEMA’s works, and that it was undisputed Suno obtained its training data by stream ripping (¶ 11). It cites a press report for that ruling. It also mentions suits by Denmark’s Koda and by Canada’s SOCAN (¶ 11 n.12).

On fair use, the complaint relies on Bartz v. Anthropic for the rule that keeping copies indefinitely in a general-purpose library is not fair use (¶¶ 68, 76). It also cites the Third Circuit’s 29 September 2026 decision in Thomson Reuters v. ROSS Intelligence, which it describes as rejecting fair use for AI training that competes with the source product (¶ 72).

Damages math and remedies

Each row total is stated in the complaint (¶¶ 83, 88, 94). The sum is our arithmetic: $157,500,000 + $2,625,000 + $26,250,000 = $186,375,000. The complaint does not add the counts together. The DMCA figures assume one violation per Work. The infringement figure is pleaded as an alternative to actual damages and Suno’s profits (¶ 83). The prayer also asks for impoundment or destruction of copies in training datasets, repositories, model weights and servers under § 503(b), and an accounting of all training data and scraping involving the Works (Prayer ¶¶ (e)-(f)).

Related matters

• Reservoir Media Management, Inc. v. Anthropic PBC, No. 5:26-cv-12308 (N.D. Cal., filed 21 Oct 2026), a parallel suit by the same plaintiffs.
• Round Hill Music LP v. Suno, Inc., No. 3:26-cv-8507-RFL (N.D. Cal.), and Gerencia 360 Music, Inc. v. Suno, Inc., No. 3:26-cv-09248-RFL (N.D. Cal.), which the complaint cites for output examples (¶ 50).
• UMG Recordings, Inc. v. Suno, Inc., No. 1:24-cv-11611-FDS (D. Mass., filed 24 June 2024), the record-label suit the complaint cites for Suno’s answer on training data and for output examples (¶¶ 32, 56-60).