The songs named in the complaint include Eye of the Tiger, Hallelujah, September, Uptown Funk, Taylor Swift's Paper Rings, and Mariah Carey's All I Want for Christmas Is You.
There is something faintly absurd about a federal filing that reads like a wedding playlist, and the absurdity wears off quickly once you get to the numbers. Tens of thousands of compositions, at up to $150,000 each in statutory damages, is arithmetic that runs into the billions.
But the number is not the interesting part, and neither is the music. What makes this case worth ten minutes of a small business owner's attention is a procedural choice: the publishers did not only sue the company. They sued two individual people.
What was actually filed
Sony Music Publishing, Warner Chappell, and several other publishers filed in the US District Court for the Northern District of California in late August 2026. The allegation is what they describe as a brazen campaign of illegally torrenting, scraping, and downloading copyrighted works.
Specifically, the complaint alleges Anthropic obtained lyrics and sheet music through pirate sources including Library Genesis and the Pirate Library Mirror, and separately scraped licensed lyric sites such as Musixmatch and LyricFind. The publishers claim this material went into Claude's training data and that the model can reproduce copyrighted lyrics on request.
The suit brings four counts, and their structure is deliberate. Direct infringement by torrenting is brought against all three defendants, meaning the company and both founders. Contributory infringement by torrenting is brought against Amodei and Mann specifically. The remaining two counts, direct infringement and removal or alteration of copyright management information, are brought against Anthropic alone.
Anthropic's response has been brief and unambiguous: it disagrees with the claims and intends to defend itself robustly in court. Nothing has been decided, and it is worth holding that firmly, because a complaint is a set of allegations rather than a finding.
Why naming the founders matters
Companies get sued constantly. It is a cost of operating, it is what corporate structure exists to absorb, and most business owners have learned to read corporate litigation as background noise.
Naming individuals is a different move. It signals that the plaintiffs believe specific people made specific decisions that fall outside the protection a company normally provides, and it is chosen deliberately because it raises the stakes for the defence in a way that suing an entity does not. A company can settle and move on. A person named personally carries the outcome with them.
The legal theory behind the contributory infringement count against Amodei and Mann is essentially that they knew about and facilitated the conduct. Whether that survives contact with a court is genuinely unknown and worth watching, because if courts prove receptive to it, the calculation changes for anyone making decisions about training data anywhere in the industry.
For a small business owner the transferable principle is narrow but real. Corporate structure protects you from a great deal, and it protects you least when a plaintiff can argue you personally directed conduct you knew was improper. That is not new law and it is not usually top of mind, and this case is a reminder that the shield has a shape rather than being absolute.
The distinction the last case established
This is not Anthropic's first copyright case, and the outcome of the previous one explains why this complaint is built the way it is.
In the authors' class action that Anthropic settled for $1.5 billion, the court drew a line that has shaped everything since. Incorporating copyrighted works into AI training was not itself found to violate the law. Obtaining those works through piracy was. The training was defensible and the acquisition was not, which is a distinction most coverage flattened at the time and which is doing enormous work here.
Read the Sony and Warner complaint with that in mind and its architecture becomes obvious. The emphasis is overwhelmingly on how the material was obtained: torrenting, Library Genesis, the Pirate Library Mirror, scraping licensed sites. The publishers are aiming squarely at the part of the previous ruling that went against Anthropic rather than relitigating the part that did not. We went through the earlier rulings in what AI copyright decisions mean for small business, and this filing is the direct descendant of them.
That distinction also explains why this is not simply a repeat. The $1.5 billion settlement covered books and authors. This covers musical compositions and publishers, a different rights holder group with different licensing infrastructure and, in the music industry's case, considerably more experience litigating aggressively about it.
What is not your problem
It is worth being clear about the limits of this, because AI copyright coverage tends toward a vague dread that is not actionable and not proportionate.
You are not liable for what a model provider used to train its model. If you use Claude, ChatGPT, or Gemini in your business, the question of how those models were trained is a dispute between the provider and the rights holders. Nobody is coming for a bakery in Aarhus because Anthropic may have torrented sheet music. That is not how this works and it is not a risk worth carrying around.
You are also not exposed by the ordinary outputs of these tools. Text a model generates for your marketing, summaries it writes of your own documents, replies it drafts to your customers: none of that is implicated by a case about training data acquisition. The vast majority of business AI usage sits comfortably outside anything this litigation touches.
And the outcome, whenever it arrives, will most likely change your costs rather than your legality. If the industry ends up licensing music and text at scale, that expense lands in the price of the tools you use. That is a budgeting consideration on a multi-year horizon, not a reason to reconsider whether to use AI at all.
What might be
There are two narrow places where this genuinely reaches a small business, and they are worth checking rather than worrying about.
The first is reproduction rather than generation. The publishers allege Claude can reproduce copyrighted lyrics on request, and that is the boundary that matters for you. A model writing original marketing copy is one thing. A model reproducing a substantial chunk of someone else's protected work, which you then publish on your website or put in a customer-facing product, is a different thing, and the fact that a machine produced it does not help you. If you have ever asked an AI for song lyrics, poetry, or extended passages from a book and used the result commercially, that is worth reviewing.
The second is what you have told clients. If you run an agency, a studio, or any business delivering creative work, your contracts may contain warranties about originality and non-infringement that you signed without thinking about AI at all. An indemnity clause promising that deliverables do not infringe third-party rights is a promise you are making, and it does not have an AI exception written into it. That is worth reading before it matters rather than after.
Both of those are review tasks rather than emergencies, and both are cheap to resolve. The common failure is not doing either because the whole topic feels too large and legal to engage with, which is how a fifteen-minute check turns into an unexamined liability.
What to actually do
The honest answer is very little, done deliberately.
Check your provider's indemnification terms. The major AI companies now offer some form of copyright indemnity for enterprise and business customers, covering you if their output triggers a claim. Coverage varies considerably between providers and between tiers, and a great many businesses are on a plan that includes it without knowing, or on one that does not while assuming they are covered. Five minutes in your account terms answers this permanently.
Then apply the one rule that covers most of the real risk: do not publish AI output that reproduces recognisable protected work. Not lyrics, not poetry, not extended passages from books, not character likenesses. This sounds obvious and it is violated regularly by businesses who asked an AI for something creative and did not consider where the creativity came from.
And keep some perspective on the timeline. This case will take years. The previous one took roughly two before settling. Nothing about your operations needs to change this quarter because of a complaint filed last month, and the businesses that handle this well are the ones that check their contracts once, set a sensible publishing rule, and then stop thinking about it until something actually happens.
Sources
- TechCrunch: Sony Music, Warner sue Anthropic, alleging a brazen campaign of intellectual property theft
- Fortune: Sony and Warner Music sue Anthropic over alleged theft of tens of thousands of songs
- Music Business Worldwide: Sony Music Publishing and Warner Chappell sue Anthropic in multi-billion dollar lawsuit
- Variety: Sony Music Publishing, Warner Chappell allege Anthropic launched brazen campaign to illegally train Claude