Major music publishing companies have escalated their confrontation with artificial intelligence companies, after a number of prominent music publishers, including two subsidiaries of Sony Music Publishing and Warner Chappell Music, filed a lawsuit against Anthropic, accusing it of illegally obtaining tens of thousands of musical compositions and using them to train the artificial intelligence model Claude.
Anthropic rejected these accusations, stressing that it will defend itself vigorously before the judiciary.
The lawsuit comes in light of the escalating dispute between the music industry and artificial intelligence developers, with intellectual property rights holders seeking to oblige companies to obtain licenses to use their works, ensuring their approval and obtaining financial compensation.
Accusations of collecting pirated content
The lawsuit did not target Anthropic alone, but also included CEO and co-founder Dario Amodei and co-founder Benjamin Mann, as plaintiffs accuse them of participating in a large operation to collect copyrighted works via torrent networks and the Internet, and later using them in the development of Claude.
The rights in dispute relate primarily to musical compositions, including song lyrics, melodies and musical notes, not just the final audio recordings.
According to prosecutors, Mann downloaded more than 5 million pirated books using the BitTorrent protocol, while Anthropic employees obtained about 2 million additional books from the Pirate Library Mirror.
Music publishers also accused the company of collecting song lyrics from licensed services, including Musixmatch and LyricFind, without obtaining permission to use them to train artificial intelligence models.
The list of works mentioned in the lawsuit includes a number of famous songs, including “All I Want for Christmas Is You,” “Eye of the Tiger,” “Livin’ on a Prayer,” “Hallelujah,” and “Uptown Funk,” in addition to the song “Paper Rings” by Taylor Swift.
Accusations of reproducing song lyrics
The accusations are not limited to the way the training data was collected, as music publishers say that Claude is able to produce parts of song lyrics that are identical or very similar to the original works.
They also claim that measures put in place by Anthropic to prevent the model from producing protected content could be bypassed by rewording the orders given to it.
The plaintiffs requested that the case be referred to a jury, and that Anthropic be ordered to pay damages of up to $150,000 for each work found to have been intentionally infringed, in addition to up to $25,000 for each incident related to the removal of copyright management information.
They also demand that the company disclose the sources of training data and data used, and destroy illegally obtained copies.
The issue of books paves the way for a new confrontation
The lawsuit is based in part on a case Bartzwhich a number of authors sued Anthropic for using their books to train artificial intelligence models.
In June 2025, a US judge distinguished between the use of books to train models and the way the company obtained them. He considered that training the model using legally obtained copies could, in certain circumstances, fall within the principle of fair use, but he believed that creating a permanent library of pirated copies would not become legal simply because they were later used in developing artificial intelligence.
Subsequently, Anthropic reached a $1.5 billion settlement with the authors.
Music publishers are using this case to argue that the dispute is not just about the legality of training AI models, but also the source of the data and the way the companies obtained it. Even if the court deemed the training process to be transformative use, obtaining pirated copies could, according to the plaintiffs, constitute a separate copyright infringement.
A previous confrontation between Anthropic and the music industry
The current lawsuit is not the first confrontation between Anthropic and the music industry, as Universal Music Publishing Group, Concord Music Group, and ABKCO previously filed a lawsuit against the company, accusing Claude of reproducing copyrighted song lyrics without a license.
In July 2026, the companies filed an amended complaint that focused on Claude’s ability to generate words that were identical or very close to the original works, as well as the effectiveness of the controls Anthropic had in place to prevent this.
But the new lawsuit expands to include other aspects, most notably the sources of training data and the method of obtaining, storing and using it. It also targets Amudi and Man personally in some accusations related to decisions to collect content.
Anthropic is not the only company
Other AI companies are facing similar lawsuits from the music industry, including song generation platform Suno, which was sued in June 2024 by record labels Universal Music Group, Sony Music Entertainment, and Warner Music Group, accusing it of using copyrighted recordings to train its models without a license.
Suno defends its practices by saying that training its models represents transformative use that falls within the doctrine of fair use. It has also requested that Sony and Universal be prevented from expanding the scope of the case to include more than 61,000 additional recordings, and has sought to keep some details of the training data confidential under the pretext of protecting its competitive information.
In November 2025, Warner exited the case after reaching a settlement and licensing partnership with Suno, while Sony and Universal continued court proceedings.
These issues reflect the mounting legal pressure on artificial intelligence companies to disclose their data sources, and the trend towards training models that rely on clear licenses and compensation for rights holders, rather than using content collected from the Internet without the consent of its owners.
