NextFin News - Sony Music Publishing and Warner Chappell Music sued Anthropic and its two co-founders on Friday, accusing the maker of Claude of running "one of the largest and most blatant ongoing thefts of intellectual property in history." The 48-page complaint, filed August 28 in the U.S. District Court for the Northern District of California, seeks statutory damages of up to $150,000 per willfully infringed work plus up to $25,000 for each instance of stripped copyright-management information — exposure the publishers say runs into the billions across tens of thousands of compositions. The case is not a fair-use argument about training AI on lyrics. It is a piracy argument about how those lyrics reached Anthropic's servers, and it lands squarely in the path of an initial public offering that the complaint says values the company at $2 trillion.
The Filing
The complaint names three defendants: Anthropic PBC, co-founder and chief executive Dario Amodei, and co-founder Benjamin Mann. It was lodged in the San Jose division of the Northern District of California, with a jury trial demanded and four counts pleaded.
With Sony Music Publishing and Warner Chappell Music now in the fray, the publishing arms of all three major music companies are litigating against the maker of Claude. Universal Music Publishing Group, Concord Music Group and ABKCO sued in Nashville in October 2023 over roughly 500 songs, in a case later transferred to California. The same publishers filed a second suit in January 2026 covering more than 20,000 works and seeking more than $3 billion. BMG brought a third case in March 2026 over 493 compositions. And on August 17, Round Hill Music filed a fourth suit against Suno and Anthropic, seeking up to $1 billion apiece.
The four counts are: direct infringement by torrenting, against all three defendants; contributory infringement by torrenting, against Amodei and Mann individually; direct infringement, against Anthropic alone; and removal or alteration of copyright management information, against Anthropic alone. Beyond damages, the publishers demand a permanent injunction, an accounting of Claude's training data and methods, and destruction of all infringing copies under court supervision.
The works named in the filing span decades and genres: "Ain't No Mountain High Enough," "All I Want for Christmas Is You," "Eye of the Tiger," "Here Comes Santa Claus," "Paper Rings," "Livin' On a Prayer," "September," "Great Balls of Fire," "Ramblin' Man," "Hallelujah" and "Uptown Funk."
The Piracy Line, Not the Fair-Use Line
The publishers' legal theory is narrower — and, they believe, stronger — than a broad assault on AI training. It targets acquisition.
The complaint alleges that in June 2021, Mann used the BitTorrent protocol to download at least five million pirated books from Library Genesis, known as LibGen, and that Anthropic employees torrented at least two million more from the Pirate Library Mirror in July 2022. Both figures are drawn from findings in Bartz v. Anthropic, the authors' class action in which a judge in the same district described Anthropic's conduct as "straightforward piracy but at massive scale." Anthropic agreed to settle that case for $1.5 billion in September 2025.
"Dr. Amodei and Mr. Mann are personally liable for their respective roles in this illegal torrenting of pirated copies of Music Publishers' works from LibGen and PiLiMi."
The filing cites internal Anthropic material unsealed in Bartz, including Mann's description of LibGen as "sketchy AF" and the company's own Archive Team describing LibGen as a "blatant violation of copyright." Beyond torrenting, the publishers allege Anthropic scraped lyrics from licensed sites including MusixMatch and LyricFind.
The publishers frame the conduct as a betrayal of Anthropic's public positioning: "Despite branding itself as the 'ethical AI company,' Anthropic has repeatedly acted in ways that belie that image, prioritizing competitive advantage over compliance with the law." And they preempt the fair-use defense by pointing to their own licensing activity: "Music Publishers recognize the potential of ethical AI technology, and they have entered licenses permitting the authorized use of their musical compositions in connection with AI."
That last sentence is the hinge of the case. A market for licensed AI training data already exists; the publishers' argument is that Anthropic chose the unpaid route.
How the Lyrics Reach the Model
The mechanism the complaint describes has two stages, and the distinction matters for both liability and damages.
First, ingestion. Books containing lyrics and sheet music enter Anthropic's training corpus through torrenting or scraping. At this stage the copyrighted work is copied in full onto Anthropic's infrastructure — a reproduction that is straightforwardly infringing unless licensed, regardless of any later fair-use argument about training.
Second, use. The ingested text is tokenized and mixed into training batches, shaping the model's weights. When a user prompts Claude for lyrics, the model generates output that, the publishers allege, reproduces protected works. The complaint's CMI count turns on this second stage: if copyright management information was stripped during ingestion or preprocessing, that is a separate violation under the Digital Millennium Copyright Act, assessed per violation rather than per work.
This two-stage structure is why the publishers can plead four counts from one course of conduct. The torrenting counts attach to the first stage. The CMI count attaches to the second. And the direct-infringement count against Anthropic alone captures the company's use of the works in training and output generation.
The Damages Arithmetic
Statutory damages under 17 U.S.C. § 504(c) reach $150,000 per work where infringement is found willful. For removal or alteration of copyright management information, the publishers seek up to $25,000 per violation under 17 U.S.C. § 1203(c)(3)(B). Alternatively, they seek actual damages plus Anthropic's profits from the infringement.
The complaint characterizes the infringed works as "thousands upon thousands" of musical compositions and describes the alleged infringement as spanning "tens of thousands" of compositions. The arithmetic is unforgiving at the top of the range: 10,000 works would imply $1.5 billion in statutory damages; 20,000 works would imply $3 billion — before any copyright-management penalties, which are assessed per violation rather than per work. The filing notes that Anthropic has "parlayed such mass infringement into a staggering $2-trillion-dollar valuation," citing an August 2026 report that the company is eyeing an October IPO at a record-breaking $2 trillion.
Even a verdict well below the ceiling would rank among the largest copyright judgments ever recorded. The publishers are not hiding their ambition; the complaint opens by stating its purpose is "to hold accountable the culprits behind one of the largest and most blatant ongoing thefts of intellectual property in history."
Why the Founders Are Named
Naming Amodei and Mann as individual defendants is unusual in copyright litigation and carries deliberate strategic weight. Personal liability pierces the corporate shield that would otherwise contain any judgment within Anthropic PBC, and it places the founders' own conduct — the alleged June 2021 BitTorrent downloads — directly before a jury.
It also signals that the publishers are not merely seeking a licensing settlement. The same firm, Oppenheim + Zebrak LLP, leads the Concord/Universal Music Group case against Anthropic, and the pattern across all four pending suits is consistent: the battleground is acquisition, not training. Contributory infringement requires knowledge and material contribution; by naming the founders, the publishers put Anthropic's internal awareness of LibGen's illegality — captured in the company's own unsealed messages — at the center of the liability question.
Cyclical Dispute, Structural Shift
This lawsuit is cyclical in one sense and structural in another, and confusing the two is the easiest way to misread it.
The cyclical layer is the litigation cycle itself. AI companies have faced a wave of copyright suits since 2023 — from authors and publishers, from music companies, from visual artists — and the pattern has been consistent: file, fight discovery, settle at a nine-figure or low-ten-figure number, repeat. Anthropic's $1.5 billion Bartz settlement fits that cycle. On this reading, the Sony and Warner case is another turn of the same wheel, and the market should expect a settlement rather than a precedent-setting verdict.
The structural layer runs deeper. What is new here is not the filing but the existence of a functioning licensing market for AI training data. Music publishers point to their own authorized AI licenses as proof that a paid alternative exists. If courts treat that market as the benchmark, the cost of the data layer shifts permanently from near-zero to a priced input — the way streaming services pay for catalogs and news organizations increasingly charge for text. That is a regime change in the economics of foundation-model training, not a cyclical fluctuation. It will not revert on its own, because a licensing market, once established and judicially endorsed, creates a durable price signal for every developer that follows.
The verdict on which layer dominates will come from the court's treatment of the CMI count. A ruling that model weights or training records carry copyright management information would extend the DMCA into the core of machine learning — a structural outcome. A narrow ruling that confines liability to the act of torrenting would leave the training question open and keep the cycle turning.
An IPO-Sized Stake in the Outcome
The timing places the lawsuit directly in the path of Anthropic's anticipated public listing. Amazon, which first invested in Anthropic in 2023 and has deployed roughly $8 billion, held a stake worth about $74.2 billion on paper as of an April filing — split between $42.2 billion in convertible notes and $32 billion in nonvoting preferred stock, marked against a $380 billion Series G valuation in February 2026. The company's valuation has since reached $965 billion, and Amazon booked $16.8 billion in pre-tax gains from the position in the first quarter of 2026 alone, including $12.3 billion from an upward revaluation.
Google holds roughly 14% of Anthropic in straight equity, contractually capped at 15% — worth about $135 billion at the current valuation. Anthropic has committed to spend more than $100 billion with Amazon on chips and cloud infrastructure over the next decade, and AWS sells Claude to enterprise customers through its Bedrock platform.
A multi-billion-dollar judgment, or an injunction restricting how Claude's training data can be used, would land on the S-1 registration statement that investors scrutinize ahead of what is widely expected to be a $1 trillion-plus IPO. Contingent liabilities of this size are exactly the kind of risk factor that can compress an offering price or delay a listing. The lawsuit does not need to succeed at trial to matter; it needs to survive long enough to become a disclosed overhang.
The Counter-Thesis
Anthropic's defense, foreshadowed by the Bartz settlement, is likely to draw a line between illegal acquisition and lawful training. The company's own Claude 2 documentation stated that "Claude models are trained on a proprietary mix of publicly available information from the Internet, datasets that we license from third-party businesses, and data that our users affirmatively share." If a court accepts that training on legally acquired lyrics is fair use, the publishers' case narrows to the torrented books that happened to contain lyrics and sheet music — a smaller, though still substantial, subset.
The strongest version of that defense attacks the damages theory at its foundation. Statutory damages are assessed per work. If the court finds that only a fraction of the "tens of thousands" of compositions were ingested through piracy rather than through licensed datasets, exposure compresses from the multi-billion range toward the high hundreds of millions — closer to the $1.5 billion Bartz benchmark that Anthropic already priced as the cost of doing business.
There is also a procedural escape route. The publishers' most aggressive lever is the copyright-management-information count, which turns on whether model weights and training records constitute "copyright management information" under 17 U.S.C. § 1202. That is an unsettled question, and Anthropic will argue that stripping CMI from a dataset is not the same as removing a copyright notice from a distributed copy.
The falsifying signal is concrete: if the court grants Anthropic's motion to dismiss the personal-liability counts against Amodei and Mann, or narrows the CMI count, the publishers' leverage drops sharply and a settlement near the Bartz figure becomes the base case. If both counts survive, the risk premium on any Anthropic IPO widens materially.
What Comes Next
In the short term, expect Anthropic to move to dismiss the personal-liability and CMI counts while the publishers press for discovery into Claude's training data — the very accounting the complaint demands. The same discovery fight played out in Bartz and ended with unsealed internal messages that damaged Anthropic's position. Discovery is where this case will be won or settled.
Over the medium term, the litigation will test whether the Bartz precedent — that pirating content to build a training library is illegal even where training on legally acquired text may be fair use — holds for music. The publishers are betting it does, and that a single song's separate protections for lyrics, melody and recording, split across different rights holders, gives each of them standing to sue. A ruling for the publishers would effectively force AI developers to treat training data the way streaming services treat catalogs: licensed, metered and paid.
In the long term, the structural question is who bears the cost of the data layer in the AI value chain. A loss for Anthropic would not end AI development, but it would raise the cost of acquiring training data and shift bargaining power toward rights holders — music publishers first, then book publishers, news organizations and visual artists. A win for Anthropic would leave the acquisition question unresolved for the next plaintiff, and the next.
Three scenarios frame the outcome. The base case is a settlement in the low-to-mid billions before trial, with the number keyed to how many works the court lets reach a jury. The publishers' upside case is a jury verdict validating per-work statutory damages across the full catalog, pushing exposure toward the complaint's implicit multi-billion ceiling and setting a precedent that travels to every pending AI-copyright suit. The downside case is a narrowed ruling that confines liability to the torrented subset, caps CMI exposure, and leaves the fair-use question for another day.
The signal to watch is the court's first substantive order on the motions to dismiss — specifically whether the personal-liability counts survive. That single ruling will tell the market how much of a $2 trillion valuation is exposed to the publishers' theory of the case.
This is a piracy trial, not a fair-use trial, and Anthropic's own unsealed words about LibGen are the evidence the publishers intend to put in front of a jury.
Explore more exclusive insights at nextfin.ai.
