Sony and Warner Music have filed a federal lawsuit against Anthropic, alleging the AI company scraped copyrighted song lyrics from authorized databases and trained its Claude chatbot to reproduce them verbatim. The suit, filed late Friday in Northern California federal court, names Anthropic along with co-founders Dario Amodei and Benjamin Mann, seeking damages that could potentially reach billions of dollars at $150,000 per alleged infringement.

This is the latest in a growing wave of music industry lawsuits targeting Anthropic. Universal Music Group, Concord, and ABKCO sued in 2023; BMG followed in March; and indie publisher Round Hill Music filed suit earlier this month. The sheer volume of legal action suggests the music industry views AI training practices as a critical threat to artist compensation and copyright protection.

What Makes This Lawsuit Different From Previous AI Copyright Cases?

The Sony and Warner Music suit uses a two-pronged legal strategy that distinguishes it from earlier copyright disputes. First, it leverages documents from a previous settlement: the Bartz v. Anthropic case, which resulted in Anthropic agreeing to pay $1.5 billion to book authors in 2025 after the company allegedly pirated thousands of books through torrent sites like Library Genesis and Pirate Library Mirror. While that settlement focused on books rather than lyrics, the plaintiffs appear to be using it to establish a pattern of conduct suggesting Anthropic systematically obtains copyrighted material through unauthorized means.

The second prong targets what the lawsuit describes as direct copyright infringement through scraping. The suit alleges that Anthropic extracted song lyrics from authorized lyric repositories, specifically naming MusixMatch and LyricFind as sources. The plaintiffs then claim that Claude can be prompted to reproduce these lyrics verbatim, citing examples including "Ain't No Mountain High Enough," "All I Want for Christmas is You," "Eye of the Tiger," "Here Comes Santa Claus," and "Paper Rings".

This approach mirrors the legal strategy used by the New York Times in its ongoing lawsuit against OpenAI, Microsoft, and Perplexity. Rather than debating whether training on copyrighted material constitutes fair use, the Times lawsuit focuses on the fact that users can allegedly extract verbatim, substantive sections of paywalled articles from ChatGPT. The legal argument centers on harm: if a chatbot can reproduce copyrighted content word-for-word, it potentially undermines the original publisher's business model by offering the same information for free.

How Are AI Companies Responding to Music Copyright Disputes?

Anthropic has not publicly responded to the Sony and Warner Music allegations. The company did not reply to requests for comment from Gizmodo, leaving its legal defense strategy unclear at this stage. However, the pattern of settlements and ongoing litigation suggests that AI companies may face mounting pressure to either negotiate licensing agreements with rights holders or fundamentally change how they source training data.

The broader context matters here. The music industry's aggressive legal posture reflects a fundamental concern: unlike text-based copyright disputes, which can invoke fair use arguments around commentary and criticism, reproducing song lyrics verbatim offers little legal cover. A few bars of a song are often the most recognizable and valuable part of the work, making the fair use defense particularly weak in music cases.

Steps to Understand the Legal Implications of AI Training Data

• Authorized vs. Unauthorized Sources: The lawsuit distinguishes between scraping authorized lyric repositories like MusixMatch and LyricFind versus obtaining content through torrent sites. This distinction matters legally because it shows intentional access to copyrighted material rather than incidental inclusion in broader datasets.
• Verbatim Reproduction as Evidence: The ability to extract exact song lyrics from Claude is presented as proof of infringement. This differs from cases where AI companies argue they only learned patterns and concepts, not specific copyrighted text.
• Pattern of Conduct Across Content Types: By linking the music lawsuit to the earlier book piracy settlement, plaintiffs are building a narrative that Anthropic systematically obtains copyrighted material through multiple unauthorized channels, which could influence how judges view the company's intent and practices.
• Damages Calculations: The $150,000-per-infringement figure could multiply across thousands or millions of songs, making the potential liability enormous and creating incentive for settlement negotiations rather than prolonged litigation.

The timing of this lawsuit, coming roughly a year after the Bartz settlement, suggests that music publishers may have been waiting for that case to conclude before filing their own claims. The settlement documents and Anthropic's admission of liability in the book piracy case likely provided valuable evidence for the music industry's legal teams.

What remains unclear is whether Anthropic will attempt to settle these disputes, as it did with the book authors, or fight them in court. A settlement could require the company to license music rights, pay retroactive damages, or agree to exclude certain training data sources in the future. A court victory for the plaintiffs could set precedent that makes training large language models on copyrighted music substantially more expensive or legally risky for all AI companies.