Sony and Warner Music sue Anthropic for using songs by The Beatles, Taylor Swift and Michael Jackson to train Claude

The publishing divisions of Sony Music and Warner Music have sued Anthropic in a California federal court, accusing the company of unlawfully using hundreds of song lyrics and scores to train its Claude AI models. The lawsuit, filed on Friday, alleges that Anthropic obtained that material through torrent downloads, including works by The Beatles, Taylor Swift, Michael Jackson and hundreds of other copyrighted artists.

According to the filing, Claude is able to reproduce protected lyrics 'word for word' when prompted, which the labels present as direct evidence of infringement. But the claim goes beyond mere reproduction: Sony and Warner argue that Anthropic also used those lyrics to teach Claude to 'generate large quantities' of new AI-created song lyrics, works that, the plaintiffs say, end up competing in the market as harmful substitutes for the original compositions they were trained on.

The case adds to a growing wave of litigation against artificial intelligence companies by copyright holders — authors, publishers, record labels and media outlets — challenging the use of their works to train AI systems. Universal Music Group had already sued Anthropic in 2023 and again this year on similar grounds, in proceedings that remain open. Anthropic, for its part, last year became the first AI company to reach a financial settlement in this type of dispute, paying $1.5 billion to resolve a class action brought by a group of authors.

That figure is precisely the crux of Sony and Warner's argument: in their lawsuit they state that Anthropic 'clearly views that as merely the cost of doing business', given that its business model 'remains built on copyright theft'. They add that $1.5 billion is not enough to deter such conduct at a company that has turned that mass infringement into a two-trillion-dollar valuation. The labels are now seeking damages of up to $150,000 for each infringed work, plus a court order barring Anthropic from continuing to use their catalog.

An Anthropic spokesperson responded that this is 'the third lawsuit from the same lawyers, recycling accusations already made in cases that are before the courts', and said the company will defend itself 'vigorously'. The company insists that training AI models constitutes 'fair use' of protected material, citing earlier statements by the judge handling the authors' case against Anthropic.

The case illustrates the unresolved tension between the generative AI industry and intellectual property holders: while tech companies defend training on protected data as legitimate use shielded by the fair use doctrine, record labels, publishers and authors insist it amounts to a commercial appropriation of their work without compensation, made worse when the resulting system can generate content that competes directly with the original works. With multiple simultaneous lawsuits — from UMG, from authors and now from Sony and Warner — Anthropic faces an ever-broader legal front, at a time when its very high market valuation has also become an argument used by plaintiffs to demand larger damages.