Reservoir sues Suno and Anthropic, and uses Suno's own licensing deals as evidence against it

Reservoir Media, which describes itself as an independent music company, holds a catalogue that includes Joni Mitchell's "Both Sides Now", Johnny Cash's "Ring of Fire" and Madonna's "Papa Don't Preach". On 21 Oct 2026 it sued Suno in the US District Court for the Northern District of California, and filed a parallel suit against Anthropic in the same court the same day. Between them the two complaints ask for about $338 million.

The money is not the most interesting part. Reservoir is using Suno's own licensing deals as evidence against it, and that argument could outlast this case.

What Reservoir is claiming

Against Suno, Reservoir lists 1,050 compositions it says it owns outright or exclusively controls in the US. It asks for up to $150,000 per wilfully infringed work, which comes to $157.5 million. On top of that it claims $2.625 million for circumventing access controls and $26.25 million for removing copyright management information, just over $186 million in total, as Music Business Worldwide tallies it.

The complaint says Suno used Bright Data's proxy services to scrape songs and lyrics from YouTube Music, Deezer and Genius, citing source code leaked to 404 Media. When 404 Media reported the hack, Suno said the code was mostly outdated and no longer in use. Reservoir also argues that Suno's guardrails against imitating named artists are easy to get round. Its example: a prompt for a Miles Davis-like blues track produced two instrumentals that Reservoir calls unlicensed derivatives of Davis's music. Davis's "So What" is in its catalogue. And it objects that Suno's terms let it train further on what users submit, guardrails or not.

The remedy that should worry every AI company is the request to impound or destroy unauthorised copies, including those held in training datasets and model weights.

The Anthropic suit covers 853 works whose lyrics Reservoir says were copied to train Claude. It claims $127.95 million in statutory damages plus about $23.5 million under the DMCA. The complaint relies on The Atlantic's AI Watchdog, a public tool for searching training datasets, as Digital Music News notes. Neither outlet reported a response from Suno or Anthropic to the new complaints.

Turning the licensing deals against Suno

Suno has spent the past year going legitimate. Warner settled with it in November 2025, BMG and Believe followed with deals, and on 9 Sep 2026 Suno launched v6, developed with "industry partners, including Warner Music Group, BMG and Believe". It retired its older models at the same time. Suno's product chief told Music Business Worldwide that v6 was trained from scratch and that its data does not include Universal or Sony recordings.

Reservoir's complaint turns that history around. By striking those deals, it says, Suno has "all but admitted that there is a functioning market" for licensing music as training data. That matters because the fourth fair-use factor asks whether a use harms the market for the work. If a working market for training licences exists, taking the material without paying looks less like fair use and more like skipping the bill. Each deal Suno signs strengthens that argument for everyone who hasn't signed one.

The US Copyright Office has said something similar. Its May 2025 report on generative AI training concludes that where licensing markets can meet AI training needs, unlicensed uses "will be disfavored" under that fourth factor.

The same court has heard this argument before

Reservoir's complaint cites Judge William Alsup's June 2025 ruling in Bartz v. Anthropic, from the same district, for the point that keeping a permanent stockpile of unlicensed copies isn't fair use. That part is fair: Alsup refused to treat Anthropic's library of pirated books as training copies and sent them to trial.

But the same order held that training itself was fair use. It also took on the market argument directly. Alsup assumed, for the purposes of the motion, that a market for training licences could develop, and held that it is still not a market the Copyright Act entitles authors to exploit.

So the licensing-market theory isn't a sure thing even in the district where Reservoir chose to file. Bartz concerned books and a text model, not songs and a generator that produces competing tracks, and one district judge's ruling doesn't bind another. Still, Reservoir's headline argument has already lost once in this court, while its piracy theory (scraping, and keeping what was scraped) stands on firmer ground.

Where artists stand

Reservoir owns and administers publishing rights. If it wins or settles, the money goes to Reservoir first and reaches writers under their existing publishing contracts, most of them signed long before anyone priced training data. Suno's licensed deals work the same way: its product chief told Music Business Worldwide that a share of revenue goes to partners and that rights holders decide how to pass it on. Digital Music News reported what looks to be an authentic internal email in which Warner chief executive Robert Kyncl framed the v6 launch as sharing revenue from Suno's paying subscribers with Warner's artists and songwriters. He didn't say on what terms.

The Copyright Office report also notes that direct licensing works best for large, centralised catalogues, such as major record companies, and is harder where ownership is spread out. Self-releasing writers and unsigned producers have no catalogue owner negotiating for them.

There is a creative cost too. A licensed dataset that is smaller and skews mainstream may narrow what a model can do, which names a real tension: the cleaner the training data, the narrower it may be. That is our inference, not a measurement.

What we think

The courtroom question is moving from "is training fair use?" to "what is the going rate?" And Reservoir's market argument cuts both ways. It is a weapon on liability, but once a market rate exists it also becomes a yardstick for damages. Reservoir asks for up to $150,000 a work. A defendant facing a proven licensing market will argue that the real loss is the licence fee it didn't pay. Read that way, the likely endgame of this wave of lawsuits is licensing at a price, not shutting the services down.

That makes the split, not the verdict, the number writers should watch. A publisher can win a training rate on a song while the songwriter's share is set by an old contract that never mentions AI. Writers in Reservoir's catalogue, and anyone else with a publisher, should ask now how AI-training income will be classified and split, before the first payment rather than after. We'll be watching whether any of this money reaches the people who wrote "Both Sides Now", and not just the company that owns it.

For more on this storyline, see our coverage of Suno's licensed v6 era, the Munich court's GEMA ruling and the class action led by Jason Isbell.

Sources

• Reservoir v. Suno, complaint, No. 3:26-cv-12305, N.D. Cal., 21 Oct 2026 (primary)
• Reservoir v. Anthropic, complaint, No. 5:26-cv-12308, N.D. Cal., 21 Oct 2026 (primary)
• Reservoir sues Suno for infringement over 1,050 songs, Music Business Worldwide
• Reservoir fires off infringement and DMCA suits against Suno and Anthropic, Digital Music News
• Introducing v6, Suno (primary)
• Suno launches v6 AI music models in partnership with WMG, BMG, and Believe, Music Business Worldwide
• Suno turns a legit leaf with v6, Digital Music News
• Bartz v. Anthropic PBC, Order on Fair Use, N.D. Cal., 23 June 2025 (primary)
• Copyright and Artificial Intelligence, Part 3: Generative AI Training, US Copyright Office, May 2025 (primary)