Round Hill Music filed two copyright complaints last week, one against Suno and data-scraping firm Bright Data, the other against Anthropic. Both are in the US District Court for the Northern District of California.
MBW reported the headline terms: statutory damages of up to USD $150,000 per work, a total Round Hill says would run to hundreds of millions and could approach or exceed $1 billion in each case, and a stated intention to take both to trial.
The Suno filing runs to 24 pages, the Anthropic one to 26. Here is what else is in them.
1. The remedy sought includes impoundment of model weights
Both prayers ask the court to order the defendants to deliver every unauthorized copy of Round Hill‘s works for impoundment or destruction, citing 17 U.S.C. § 503.
The filings define that to cover “copies retained in training datasets, internal repositories, model weights, and server infrastructure,” wording that reaches the trained models, not just the source files behind them.
Each Round Hill complaint also asks for a complete accounting of training data, scraping activity, and datasets involving the works.
The DMCA counts stack on top of the per-work figure: under § 1203, up to $2,500 for each act of circumvention and up to $25,000 for each removal of copyright management information.
For comparison, Anthropic‘s $1.5 billion settlement with book authors in September 2025 committed it to destroying the original files it torrented from Library Genesis and Pirate Library Mirror, and copies originating from them – but not the models themselves. Anthropic certified that no commercially released model had been trained on those datasets.
2. Anthropic’s own court record is the template: Suno is being measured against it
Both Round Hill complaints rest on Bartz v. Anthropic, the authors’ case in the same district.
Each quotes Judge William Alsup‘s June 2025 ruling: “There is no carveout, however, from the Copyright Act for AI companies.”
The Suno filing applies that holding to a company that was not a party to it, alleging “Suno has engaged in the exact same conduct” in retaining unlicensed copies indefinitely.
The Anthropic complaint sources its piracy account to the same case, citing torrenting from Library Genesis and Pirate Library Mirror (PiLiMi), and quotes what it says was a co-founder’s message to colleagues once PiLiMi could be torrented: “[J]ust in time!”
It further alleges that Concord II, the publishers’ second suit filed on January 28, 2026, revealed that Anthropic “had concealed its torrenting during discovery in Concord I,” a claim the publishers first raised in August 2025.
3. Round Hill quotes Anthropic’s own published dataset back at it
Anthropic released a human-feedback training dataset, hh-rlhf, on Hugging Face in 2022.
Round Hill alleges the dataset shows Claude returning copyrighted lyrics during testing, including a response to a Disney songs prompt that quoted from Let It Go.
In a second example, the complaint says, a user asked Claude to build a short story from the lyrics to Eleanor Rigby, and the model did so – while the response logged as rejected instead pointed the user to an article on writing from song lyrics.
“Anthropic deliberately trained Claude to commit copyright infringement,” the complaint says of that exchange.
Neither song is a Round Hill work. Both are cited to argue Claude reproduces copyrighted lyrics generally.
In its section arguing fair use does not apply, the filing reproduces an exchange on X over Moonshot AI‘s alleged distillation of Anthropic‘s Fable model.
Michael Kratsios, director of the White House Office of Science and Technology Policy, wrote that “large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable.”
Sarah Heck, Anthropic‘s head of public policy, replied: “Illicit, adversarial distillation is IP theft and industrial espionage that supports adversary military and intelligence capabilities.”
4. The DMCA case turns on which extraction tool was chosen
The Anthropic complaint alleges the company tested at least three text-extraction programs before ingesting data: Readability, Newspaper and jusText.
It claims jusText was ruled out because it left copyright notices and ownership details intact, which the filing says Anthropic treated as “boilerplate” and “useless junk.”
Newspaper was selected instead, according to Round Hill, for its ability to strip that information.
Round Hill makes the same argument about page layout, saying Bright Data‘s Scraping Browser renders pages without headers and footers, where copyright management information typically sits: “Making the deliberate choice to exclude headers and footers is identical to making the deliberate choice to remove CMI.”
Both Round Hill complaints cite Stevens v. CoreLogic for the double scienter standard: removal must be intentional, and the defendant must have known, or had reasonable grounds to know, it would induce, enable, facilitate or conceal infringement.
They also cite an October 2025 Concord ruling holding similar allegations sufficient at the pleading stage.
5. Bright Data has its own number attached: $75 million
The $1 billion figure in both complaints is a projection, contingent on Round Hill amending its exhibits to cover thousands more works.
The contributory infringement count against Bright Data carries the only fixed arithmetic in either filing: 500 works at $150,000 each, for $75,000,000.
Round Hill also turns Bright Data‘s own litigation record against it, citing X Corp. v. Bright Data, in which the same court rejected its jurisdictional challenge.
The complaint puts Bright Data past $300 million in 2025 revenue, growing 50% year-over-year, citing a report by Asymmetrix that calls the figure annualized recurring revenue.
The filing alleges the services Bright Data provided to Suno “are only good for copyright infringement.”
It also draws on the hacked Suno source code reported by 404 Media on July 15, including a dataset logged as youtube_music holding over two million music clips and 113,879 hours of audio – roughly thirteen years.
What else is in the filings
The Suno complaint quotes co-founder and CEO Mikey Shulman, citing a video produced with Oracle, as saying: “Our models are taught to just continue a piece of music.”
Round Hill sets that against Suno‘s marketing of itself as generating songs from whole cloth.
The same filing alleges Suno‘s guardrails are porous, claiming a prompt naming Reba McEntire is blocked while a misspelling of her name is not.
Both complaints name ElevenLabs, Musical AI, Symphonic, Soundverse, GEMA through PLAI, GCX/Rightsify , and Troveo as evidence of a functioning licensing market they say the defendants bypassed.
But PLAI launched on July 23, under a month before these filings, and GEMA says it is built for tools that help creators make music, with generative AI licensing the separate subject of its own Suno case, which GEMA won at first instance on July 31.
The Anthropic filing treats each model release as a fresh act of copying, naming Fable 5 and Mythos 5 (both June 9, 2026), Sonnet 5 (June 30, 2026) and Opus 5 (July 24, 2026).
Exhibit A in each case lists 500 musical compositions, not sound recordings, though Round Hill asserts rights in 16,873 recordings and calls the exhibit a bellwether it will amend to cover both.
MBW has contacted Suno, Anthropic and Bright Data for comment. None of the allegations has been tested in court.Music Business Worldwide
Round Hill Music recently filed two copyright complaints in the US District Court for the Northern District of California, targeting Suno and Bright Data in one case and Anthropic in another. The complaints seek substantial statutory damages, potentially reaching hundreds of millions, possibly exceeding $1 billion in each case, with an intention to take both matters to trial.
Key Elements of the Complaints
-
Remedies Sought:
Both complaints request the court to order the defendants to turn over any unauthorized copies of Round Hill’s works for impoundment or destruction, citing 17 U.S.C. § 503. This request extends beyond just source files, encompassing model weights and training datasets. Round Hill also demands a full accounting of the training data and scraping activities related to its works. Additionally, under the DMCA, damages for circumvention of copyright management information can amount to $2,500 for each act, plus up to $25,000 for each instance of removal of such information. -
Reference to Previous Cases:
The Round Hill complaints draw parallels to the case of Bartz v. Anthropic, where Judge William Alsup ruled that there are no exemptions from the Copyright Act for AI companies. The Suno complaint alleges that the company engaged in similar unauthorized practices as Anthropic. The Anthropic filing details claims of illegal downloading from Library Genesis and Pirate Library Mirror, suggesting that Anthropic concealed this during legal discovery. -
Use of Anthropic’s Own Data Against It:
Round Hill highlights Anthropic’s own training dataset, hh-rlhf, which allegedly includes copyrighted lyrics, to argue that Anthropic trained its model, Claude, to infringe copyrights. This claim is bolstered by specific examples of Claude reproducing copyrighted lyrics when prompted. -
DMCA Violations and Extraction Tools:
The complaints focus on the methods used by Anthropic to extract training data, alleging that deliberate choices were made to avoid retaining copyright notices. The use of the Newspaper extraction tool, which strips away such information, is cited as evidence of intentional infringement. Round Hill argues that similar methodologies were used by Bright Data, which allegedly facilitated copyright infringement through its services. -
Financial Implications:
The complaint against Bright Data specifies a potential $75 million liability based on 500 works at $150,000 each. Round Hill argues that Bright Data’s revenue, reportedly over $300 million in 2025, indicates it has profited significantly from its services that allegedly infringe copyright.
Additional Insights from the Filings
- The filings provide detailed accounts of Suno’s marketing practices and its claims about generating music from scratch. They also highlight potential weaknesses in Suno’s copyright protections, suggesting that its guardrails are insufficient to prevent infringement.
- Both complaints reference a number of other companies in the music licensing space to illustrate a functioning market that the defendants are bypassing.
- The Anthropic filing considers each model release as a new act of copying, listing specific model names and release dates to strengthen its case against the company.
Conclusion
The allegations put forth by Round Hill Music in these copyright complaints reflect a growing tension between the music industry and AI companies utilizing copyrighted materials without authorization. If successful, these cases could set significant precedents regarding the liabilities of AI companies under copyright law. The outcomes may also influence the operational landscape for data scraping firms like Bright Data and generative AI companies like Suno and Anthropic, potentially reshaping how they approach copyright issues in their technological frameworks. As of now, none of these allegations have been legally validated, and the cases are poised for further developments in court.
