Sony and Warner Chappell sue Anthropic over 'tens of thousands' of songs

Sony Music Publishing and Warner Chappell have filed a multibillion-dollar lawsuit against Anthropic, the startup behind the Claude chatbot, alleging the company misused 'tens of thousands' of copyrighted songs to train its AI models. The music publishers, which manage copyrights on behalf of songwriters and composers, are seeking damages for what they describe as systematic copying of lyrics and other musical works without a license.

The lawsuit, reported by The Guardian, centers on Anthropic's training data, the vast corpus of text and lyrics that teaches Claude to generate responses. Publishers argue that feeding copyrighted songs into that corpus, and reproducing them in outputs, is not fair use but plain theft. The case is likely to turn on whether AI training counts as transformative use under copyright law, a question courts have only begun to answer.

Anthropic is not alone in the crosshairs. AI companies have already faced lawsuits from authors, visual artists, and news outlets over the use of copyrighted material in training data. What makes this suit notable is the scale: 'tens of thousands' of works, many of them chart-topping hits, and a damages figure that could run into the billions if the publishers prove willful infringement. For AI companies, the math is simple. Licensing music at scale would be expensive, but losing a multibillion-dollar judgment would be far worse.

Sony Music Publishing and Warner Chappell are two of the largest music publishers in the world, controlling catalogs that include some of the most recognizable songs of the past century. They license lyrics and compositions for everything from streaming services to karaoke machines, and they have been aggressive in policing unauthorized use. Their move against Anthropic signals that the music industry sees AI models as a new kind of unlicensed platform, one that can reproduce lyrics on demand and, in some cases, generate melodies that echo existing works.

For Anthropic, the lawsuit lands at a delicate moment. The company has raised billions from major tech investors and positioned itself as a safety-first AI lab, but its commercial success depends on Claude being useful, and usefulness requires training on massive amounts of text. If courts rule that unlicensed training data is infringement, AI labs will need to either strike licensing deals with content owners, build datasets from public-domain or openly licensed sources, or pay damages retroactively. Each option changes the unit economics of AI development.

Executives across tech, media, and entertainment should watch this case closely. A ruling against Anthropic could set a precedent that forces every AI company to audit its training data and negotiate with rights holders. It could also embolden other industries, from news publishers to book authors, to bring similar claims. Conversely, a ruling in Anthropic's favor would give AI labs a powerful fair-use shield, potentially devaluing the very content licenses that media companies are trying to sell.

The lawsuit also highlights a growing divide between AI companies and creative industries. While some publishers have chosen to partner with AI labs, others are choosing litigation. Music publishers may be betting that a courtroom victory is worth more than any licensing fee Anthropic would offer. For boards and CEOs, the takeaway is that intellectual property is no longer a back-office concern. It is a front-line strategic risk that can determine whether an AI product launches on time, at what cost, and under what legal exposure.

Anthropic's response, when it comes, will likely argue that training on copyrighted material is transformative and that Claude does not meaningfully compete with the original songs. But the publishers' claim that 'tens of thousands' of works were used without permission is a direct challenge to the industry's core assumption that public data is free to mine. Until a court decides, every AI company with a language model is effectively on notice: your training data may be your biggest liability.