Sony and Warner Music Sue Anthropic Over AI Training on Copyrighted Lyrics

A coalition of major music publishers, including Sony Music Publishing and Warner Chappell Music, has launched a sweeping copyright infringement lawsuit against artificial intelligence developer Anthropic. The legal action accuses the San Francisco-based AI firm of engaging in the "blatant theft" of copyrighted song lyrics to train its popular Claude family of large language models.[3][4]

The lawsuit, filed in federal court, represents a significant escalation in the ongoing tension between the entertainment industry and the generative AI sector. Publishers allege that Anthropic systematically scraped the internet for vast quantities of text, indiscriminately ingesting protected musical compositions without seeking permission or offering compensation to the rights holders.[1][2]

At the heart of the complaint is the assertion that Anthropic's models not only learned from these copyrighted works but can actively reproduce them. According to the filing, users interacting with Claude can prompt the AI to generate identical or nearly identical copies of famous song lyrics, effectively turning the chatbot into an unlicensed lyric aggregator.[2][5]

This capability, the publishers argue, directly undermines existing licensing markets. Companies that legally aggregate and display lyrics pay substantial fees to music publishers for those rights. By providing the same service for free through its AI interface, Anthropic is allegedly bypassing the established economic framework that supports songwriters and composers.[4][5]

Anthropic, which recently secured a multi-billion dollar valuation and positions itself as a safety-focused AI developer, now faces the same legal scrutiny that has targeted competitors like OpenAI. While the company has not yet issued a formal response to this specific filing, the AI industry at large has consistently defended the practice of scraping publicly available data as "fair use" under US copyright law.[1][3]

The fair use defense hinges on the argument that training an AI model is a transformative act. Tech companies maintain that their models do not store exact copies of the training data in a database, but rather analyze the statistical relationships between words to learn language patterns and generate novel responses.[2]

However, the music publishers' lawsuit directly challenges this premise by highlighting instances where Claude outputs verbatim lyrics. Legal experts note that demonstrating a model's ability to memorize and regurgitate protected works significantly weakens the transformative use argument, making this a particularly dangerous legal vulnerability for AI developers.[3][5]

The outcome of this litigation could fundamentally reshape the economics of artificial intelligence. If the courts rule that training AI on copyrighted material requires explicit licensing agreements, companies like Anthropic may be forced to purge their existing models and negotiate costly data access deals, shifting the balance of power back toward traditional media and entertainment conglomerates.[1][4]