Loading today’s stories
Leadership1 distinct publisher3 min readPublished
The publisher's complaints against Suno and Anthropic plead DMCA circumvention beside infringement, which moves a buyer's diligence question onto the corpus rather than the filter.
The Board Room · Leadership desk
Compiled by The Board RoomSomething wrong?How this is made
Fair use is a defense to copying. It is not obviously a defense to getting past a lock, and it has nothing to say about stripping the ownership data out of a file. That is the seam Round Hill is working. Both complaints plead, alongside direct infringement, DMCA circumvention of technical access controls and DMCA removal of copyright management information [15], and Round Hill's stated position is that its case does not turn entirely on how the fair use question resolves in the other suits [17].
The contrast with the pending Boston litigation is the whole point. There, Suno defends by arguing that training on legitimately acquired recordings is fair use [12]. Round Hill's answer is to attack the word "acquired": it alleges the training set was scraped off the open web, YouTube included [2], with an Israeli scraping firm, Bright Data, named as a third defendant for allegedly supplying the proxy networks and tools [16]. Against Anthropic the scope narrows to lyrics and compositions, since Claude generates text rather than audio, but the access-control theory is the same [3].
For anyone embedding either model, that distinction decides where the diligence goes. Round Hill does cite output as evidence, pointing to Claude rewrites that it says kept the structure and melodic hooks of the originals with little changed beyond surface wording [4]. But a filter tuned after the fact cannot unscrape a corpus. If the circumvention and CMI counts have legs, the operative facts were fixed at collection time, inside a vendor's pipeline, possibly at a subcontractor the customer has never been told about.
The arithmetic is worth sitting with. Round Hill's plaintiffs control 14,364 compositions and 16,873 recordings [6], which is 31,237 works [1]. The filings name roughly 500 songs as a bellwether, with a stated plan to reach 10,000 or more compositions [7] - so the current sample is about 3.5 percent of the composition catalogue, and the target is about 70 percent of it [2]. Each suit separately seeks damages that could exceed a billion dollars [5], against a portfolio the company values at $1.1 billion [1]; combined, the demand is roughly 1.8 times the asset it is meant to protect [3].
The exit that worked before may not be available. Warner settled with Suno in November 2025 and took a licence, and BMG never sued at all, going straight to a licensing alliance [13]. Round Hill's CEO Josh Gruss says the company will not settle [8], and its counsel, Richard Busch, who won Blurred Lines for the Marvin Gaye estate, has framed it as "there is nothing fair about that use" [9]. Suno has also already lost the no-licence-needed argument once, in Munich on 31 July 2026, in a case brought by GEMA, on German and EU law that binds no US court [10].
Ranked by verification strength, evidence, and original report placement.
Round Hill's complaint cites an example where Claude, asked to rewrite several of the songs at issue for a modern audience, produced output the company says stayed extremely close to the originals in structure and melodic hooks, changing little beyond surface wording.
Round Hill's complaints make four separate legal arguments, three of them being direct copyright infringement under the Copyright Act, DMCA circumvention of technical access controls, and DMCA removal of copyright management information, which Round Hill alleges both companies stripped out during scraping.
Round Hill Music, an independent music publisher with a $1.1 billion portfolio of music copyrights, filed two separate copyright lawsuits on August 17, 2026, both alleging the defendants trained AI models on its copyrighted music without a licence or permission.
The Suno suit covers both compositions and recordings Round Hill controls, and alleges Suno built its training dataset by scraping copyrighted music directly off the open web, including YouTube, without permission or a licence.
The Anthropic suit is narrower, covering only lyrics and compositions rather than recordings because Claude is a text model, but makes the same category of claims: training on copyrighted lyrics without a licence, with material obtained through unauthorized copying and access-control circumvention.
Each of the two suits seeks damages that could exceed $1 billion.
Follow any of these and your For You feed starts watching them — no settings page required.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Detailed pleadings, single publisher, nothing adjudicated in the US
The factual spine is unusually specific for a filing story — exact work counts, named songs, the four pleaded theories, the named co-defendant, the Boston consolidation and judge — and one external data point, the Munich GEMA ruling, is a decided outcome rather than an allegation. But everything about the two new suits is drawn from complaints reported by one publisher, no defendant response appears, no docket or schedule is given, and the core legal propositions (DMCA circumvention, CMI removal, stockpile-fails-fair-use) are untested in these cases.
Licensing has become the dominant resolution path; this suit refuses it
There is concrete, dated real-world uptake of the licensing route rather than of Round Hill's theory: Warner exited the Suno case in November 2025 for a licensing deal and settled with Udio the same week, Universal settled with Udio in October 2025, and BMG licensed directly without ever suing. Against that, only Sony and Universal (versus Suno) still litigate, and Round Hill's DMCA-and-stockpile approach has zero adopters or judicial endorsement in the US so far — its only supportive real-world outcome is a foreign, non-binding ruling.
Headline asserts a conclusion the record has not reached
The framing — that illegally scraped music 'isn't fair use' — states as settled what is currently an untested pleading in two brand-new US cases; the only ruling in hand is German and expressly non-binding. Damages 'could exceed $1 billion' per suit, a combined demand near 1.8x the plaintiff's own stated portfolio value, is presented without the statutory-damages arithmetic behind it, and the named-song list is 3.5 percent of the catalog with the 10,000-work expansion still aspirational. The underlying mechanism claims are specific and well documented, which keeps the gap moderate rather than severe.
Narrative sourced almost entirely from the plaintiff side
The load-bearing quotes and characterizations come from parties with direct financial stakes in the outcome: the plaintiff's CEO announcing he will not settle, and the plaintiff's litigator — hired for a headline copyright win — asserting 'there is nothing fair about that use' and framing the stockpile theory. The damages figures, the similarity assessment of Claude's output and the scope-expansion plan are all plaintiff assertions that also serve as settlement leverage. No defendant, judge or disinterested expert is quoted, and the article does not disclose any independent check on the plaintiff's characterizations.
Facts of the filings are firm; their consequences are not
Confidence is limited by structure rather than sloppiness: one publisher, one side, and legal theories whose fate is unknowable from the record supplied. The verifiable layer — who filed, when, against whom, on what counts, and the prior settlement and ruling history — is internally consistent and precise enough to rely on. The predictive layer, including whether the DMCA and stockpile arguments will shift liability from output to intake, cannot be assessed from this cluster.
invest
Three labels, one cap table: the majors buy into Stability AI instead of suing it3 distinct publishers
product
Alsup fined the download, not the training: the copyright risk in your stack is provenance1 distinct publisher
product
The labels bought equity, not just injunctions, in Stability AI's $76mn round3 distinct publishers
build
Your Training Data Now Has Counterparties: The Case For Treating Corpora As Procurement1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
forbes.com
1 article · August 26, 2026