Sony and Warner Sue Anthropic $150K Per Song Sought Over AI Lyric Scraping.

Sony and Warner Music File Massive Copyright Lawsuit Against Anthropic Over AI Lyric Scraping

The music publishing divisions of Sony Music Publishing and Warner Chappell Music filed a joint lawsuit against AI startup Anthropic in U.S. federal court in California. The suit alleges that Anthropic illegally scraped and used copyrighted song lyrics to train its flagship AI model, Claude, without authorization or licensing agreements.

The record labels are seeking statutory damages of up to $150,000 per infringed work, escalating legal pressure on generative AI developers over intellectual property rights.

Torrent Downloads, Verbatim Output, and $1.5B Settlement Critique

The complaint alleges that Anthropic infringed on hundreds of iconic songs by legendary artists, including The Beatles, Taylor Swift, and Michael Jackson. According to court filings, Anthropic acquired these lyrics and musical compositions by downloading torrent files containing pirated song databases to train its Claude models. The labels demonstrated that when prompted, Claude can generate near-verbatim copies of copyrighted lyrics.

This lawsuit marks an expansion of the legal front against Anthropic:

• Existing Universal Lawsuit: Universal Music Group (UMG) filed a similar copyright infringement lawsuit against Anthropic in 2023, which remains active in court.

• The $1.5 Billion Class Action: While Anthropic previously agreed to a $1.5 billion settlement in a separate author class-action lawsuit, Sony and Warner explicitly argued in their filing that a settlement of that size is insufficient to deter violations by a company operating within an AI sector backed by multi-trillion-dollar valuations.

Anthropic’s Defense: Fair Use and Repeated Claims

Anthropic firmly rejected the lawsuit, stating that this represents a third legal action brought by the same team of attorneys recycling identical allegations. The company maintains that its AI model training constitutes legal Fair Use, citing legal reasoning established in earlier author cases, and confirmed its intention to vigorously defend its technology in court.

Music publishers counter that Claude's ability to generate commercial-grade lyrics creates market substitutes that directly undermine the economic value of original copyrighted works.

How Copyright Law Evaluates generative AI to Prove Copyright Infringement Under US Law, plaintiffs must demonstrate not only that copyrighted data was copied during model training, but that the result acted as a detrimental "market substitute" that harmed the commercial value of the original creator. Publishers argue that if enterprise users could prompt Claude to write songs in a style and lyrics that matched the copyright of his hit songs, it would reduce the need for formal licensing agreements.

While AI labs often argue that copying publicly accessible web pages falls under fair use, obtaining dataset files via torrent networks creates a significant legal loophole. Courts view torrent downloads of infringing datasets as more serious than standard web data collection, making it harder for AI developers to claim good faith research practices.

The rise in lawsuits from the three major music publishers (Universal, Sony, and Warner) indicates that the music industry is attempting to force AI companies into a standardized licensing system, similar to how Spotify and Apple Music standardize streaming licensing, rather than halting AI development entirely. Record labels aim to impose licensing fees based on each song or subscription used to train AI models.

 

 

💬 AI Content Assistant

Ask me anything about this article. No data is stored for your question.

Sony and Warner Music File Massive Copyright Lawsuit Against Anthropic Over AI Lyric Scraping

The music publishing divisions of Sony Music Publishing and Warner Chappell Music filed a joint lawsuit against AI startup Anthropic in U.S. federal court in California. The suit alleges that Anthropic illegally scraped and used copyrighted song lyrics to train its flagship AI model, Claude, without authorization or licensing agreements.

The record labels are seeking statutory damages of up to $150,000 per infringed work, escalating legal pressure on generative AI developers over intellectual property rights.

Torrent Downloads, Verbatim Output, and $1.5B Settlement Critique

The complaint alleges that Anthropic infringed on hundreds of iconic songs by legendary artists, including The Beatles, Taylor Swift, and Michael Jackson. According to court filings, Anthropic acquired these lyrics and musical compositions by downloading torrent files containing pirated song databases to train its Claude models. The labels demonstrated that when prompted, Claude can generate near-verbatim copies of copyrighted lyrics.

This lawsuit marks an expansion of the legal front against Anthropic:

• Existing Universal Lawsuit: Universal Music Group (UMG) filed a similar copyright infringement lawsuit against Anthropic in 2023, which remains active in court.

• The $1.5 Billion Class Action: While Anthropic previously agreed to a $1.5 billion settlement in a separate author class-action lawsuit, Sony and Warner explicitly argued in their filing that a settlement of that size is insufficient to deter violations by a company operating within an AI sector backed by multi-trillion-dollar valuations.

Anthropic’s Defense: Fair Use and Repeated Claims

Anthropic firmly rejected the lawsuit, stating that this represents a third legal action brought by the same team of attorneys recycling identical allegations. The company maintains that its AI model training constitutes legal Fair Use, citing legal reasoning established in earlier author cases, and confirmed its intention to vigorously defend its technology in court.

Music publishers counter that Claude's ability to generate commercial-grade lyrics creates market substitutes that directly undermine the economic value of original copyrighted works.

How Copyright Law Evaluates generative AI to Prove Copyright Infringement Under US Law, plaintiffs must demonstrate not only that copyrighted data was copied during model training, but that the result acted as a detrimental "market substitute" that harmed the commercial value of the original creator. Publishers argue that if enterprise users could prompt Claude to write songs in a style and lyrics that matched the copyright of his hit songs, it would reduce the need for formal licensing agreements.

While AI labs often argue that copying publicly accessible web pages falls under fair use, obtaining dataset files via torrent networks creates a significant legal loophole. Courts view torrent downloads of infringing datasets as more serious than standard web data collection, making it harder for AI developers to claim good faith research practices.

The rise in lawsuits from the three major music publishers (Universal, Sony, and Warner) indicates that the music industry is attempting to force AI companies into a standardized licensing system, similar to how Spotify and Apple Music standardize streaming licensing, rather than halting AI development entirely. Record labels aim to impose licensing fees based on each song or subscription used to train AI models.