Sony and Warner vs. Anthropic

Marisa Shah
7 days ago
6 min read

The music industry’s next major battle over artificial intelligence is underway. Sony Music Publishing, Warner Chappell Music, and 33 affiliated music publishers have sued Anthropic, alleging that the company used pirated lyrics, sheet music, and other copyrighted musical works to build and operate its Claude AI models. The case could help determine whether AI companies can treat creative work as raw material for model training or whether they must license it, credit it, and pay for it.

The complaint was filed August 28, 2026, in the U.S. District Court for the Northern District of California. It names Anthropic PBC as well as CEO and co-founder Dario Amodei and co-founder Benjamin Mann. The publishers characterize the alleged conduct as a campaign of mass copyright infringement carried out to develop and profit from Claude.

At the center of the case are thousands of musical compositions. The complaint gives examples ranging from “Ain’t No Mountain High Enough” and “All I Want for Christmas Is You” to “Eye of the Tiger,” “Livin’ on a Prayer,” “September,” and “Hallelujah.” The plaintiffs say these examples are illustrative, not a complete accounting of the works they believe were infringed.

The publishers’ claims go beyond the idea that Claude may have been trained on copyrighted text. They allege several distinct forms of copying:

- Downloading millions of books from pirate repositories, including Library Genesis and Pirate Library Mirror, some of which contained songbooks and sheet music.

- Scraping lyrics from authorized services such as Musixmatch and LyricFind.

- Scanning secondhand books to digitize their contents.

- Acquiring material from large datasets including Common Crawl, The Pile, and Books.

- Making further copies during model training and fine-tuning.

- Producing verbatim or near-verbatim lyrics in Claude’s outputs.

- Removing copyright-management information, such as song titles, songwriter names, and other attribution data.

Sony and Warner are not merely alleging that Anthropic learned patterns from music. They contend that the company obtained the underlying materials unlawfully, made copies at multiple stages, and enabled outputs that can reproduce protected lyrical expression. That turns the litigation into a test of both training data provenance and output behavior.

Why music publishers are drawing a line

Music publishers occupy a different position from record labels. A record label generally controls a particular recorded performance, while a music publisher administers the underlying composition: the lyrics and music written by songwriters. That distinction matters because the Sony and Warner suit focuses on compositions, lyrics, and sheet music, all the material at the core of songwriting rights.

The complaint argues that lyric websites are not free merely because consumers can see lyrics online. According to the publishers, services displaying lyrics pay for licenses and generally provide songwriter and title information. From their perspective, an AI company systematically copying that material for commercial model development is not comparable to a listener reading a lyric page. It is an unlicensed industrial use of a catalog built through investment in songwriters and copyrights.

The publishers also say that legitimate AI licensing models already exist. Their complaint states that they have entered licensing agreements with AI companies for authorized uses of musical compositions. That claim is strategically important: it undercuts the argument that licensing is impossible or that AI development inherently requires uncompensated access to copyrighted works.[1]

The stakes:

Sony and Warner are seeking a jury trial, an injunction, damages, and other relief. They seek statutory damages of up to $150,000 for each work found to have been willfully infringed, plus up to $25,000 per violation involving the removal or alteration of copyright-management information. The complaint does not establish a final damages figure, but because it alleges infringement involving tens of thousands of compositions, the theoretical exposure could reach into the billions if the publishers prevail on a large scale.

An injunction may be more consequential than money. If a court orders Anthropic to stop using particular works, remove them from training libraries, or retrain or modify models, that could create costly operational consequences for the AI industry. It would also pressure AI developers to keep detailed records of where training material came from something publishers and other rights holders have increasingly demanded.

For songwriters, the case is about more than a payout. The publishers argue that unlicensed AI-generated lyrics can become substitutes for licensed music, weaken the market for lyrical works, and reduce the economic value of songwriting. The lawsuit asks the court to recognize that the ability to generate large volumes of plausible, style-adjacent lyrics can affect a market even when a specific output is not an exact copy of a known song.[1]

Anthropic’s previous copyright issues

The Sony-Warner lawsuit arrives after Anthropic’s landmark settlement with book authors and publishers. In July 2026, a federal judge gave final approval to Anthropic’s $1.5 billion settlement of a class-action case involving books used in connection with Claude, which is the largest known copyright settlement in U.S. history.

That earlier litigation produced a crucial legal distinction. Judge William Alsup held that using legally obtained books to train an AI model could qualify as fair use because the process was transformative. But the court separately found that Anthropic’s storage of more than 7 million pirated books in a central library violated copyright holders’ rights; the alleged piracy was not protected by fair use.

That distinction gives the music publishers a clear litigation strategy. Their complaint emphasizes alleged torrenting, pirate libraries, scraping, and unauthorized acquisition, not solely the abstract question of whether training on copyrighted material can be fair use. In other words, even if a court eventually accepts some form of fair-use defense for properly acquired training data, it may be far less sympathetic where the inputs were allegedly obtained from pirated sources.

Anthropic has previously maintained that training AI on books was fair use, a position it reiterated after the book settlement. But as of September 1, no public court ruling has resolved Anthropic’s response to the Sony and Warner allegations, and the complaint represents the publishers’ claims, not findings of fact.

The case also arrives at a culturally sensitive moment for the music business. AI-generated songs, voice clones, synthetic performers, and tools that mimic songwriting conventions have moved from novelty to business issue. Labels, publishers, streaming platforms, and artists are trying to distinguish between AI used as a production tool and AI trained or deployed in ways that could displace human creators without consent.

What happens next

The case is still at its opening stage. The court will first have to address procedural issues, discovery, and the parties’ factual disputes about what Anthropic obtained, how it obtained it, whether specific works were included in its data, and whether Claude can reproduce protected lyrics under present safeguards.

Several questions will likely define the case:

1. How much copyrighted music was actually acquired and used?

2. Were the materials acquired legally

3. Does Claude reproduce protected lyrics?

4. Is there a viable licensing market for AI Music Training?

5. Can executives face individual liability?

The larger meaning

The Sony-Warner case is not just a dispute over lyrics. It is a referendum on the business model behind generative AI. The publishers are arguing that scale does not convert copying into permission, and that an AI company’s technological ambitions do not erase the licensing system that supports songwriters.

Anthropic, meanwhile, is likely to rely on arguments familiar from other AI cases: model training is transformative, models do not function as searchable archives, and broad limits on training data could slow innovation. But the alleged use of pirate sources makes this lawsuit harder to frame as a clean debate about fair use. The question may be less about can AI learn from culture? than and more of what rights and obligations apply when a company industrializes access to culture without permission?

If the publishers win meaningful injunctive relief or a large damages award, AI companies may be forced toward a future of music-data licensing, audit trails, provenance controls, attribution safeguards, and stronger output restrictions. If Anthropic defeats the claims broadly, it could strengthen the position of AI developers that training on copyrighted text is generally permissible. Either way, the lawsuit has the potential to set rules that shape how the next generation of music, and the systems that generate it is made.

Sony and Warner vs. Anthropic

Recent Posts

Comments