Sony Music Publishing and Warner Chappell Music have initiated legal proceedings against Anthropic, alleging that the training of their AI model, Claude, involved the use of pirated songbooks sourced from “shadow libraries.” The claim cites approximately 7 million books, alongside various song lyrics collected from the internet.
Key Points:
- The publishers assert that Anthropic obtained copyrighted works through torrent downloads and web scraping of song lyrics.
- The complaint indicates that the 7 million downloaded books are linked to shadow libraries and included sheet music and songbooks.
- The plaintiffs are seeking up to 0,000 for each work that a jury finds was willfully infringed upon.
Torrents and Data Sources
The lawsuit was filed on August 28, 2026, in the U.S. District Court for the Northern District of California. Defendants named in the suit include Anthropic, co-founder and CEO Dario Amodei, and co-founder Benjamin Mann. Central to the allegations are the data sources used for training the AI.
According to the plaintiffs, in June 2021, Mann downloaded around 5 million books from Library Genesis, followed by an additional 2 million books from Pirate Library Mirror in July 2022. The publishers contend that these datasets contained unauthorized sheet music and songbooks featuring their protected works.
The lawsuit further claims that Anthropic scraped song lyrics from Musixmatch and LyricFind, platforms that acquire licenses for displaying lyrics. The plaintiffs describe these actions as “one of the largest and most egregious ongoing thefts of intellectual property in history.” Two of the four claims directly relate to torrent downloads.
Also of Note: Investment funds in gold have reached a six-month peak with an influx of .21 billion in a week.
Impact on Sony
The publishers are seeking statutory damages of up to 0,000 for each work that a jury determines was willfully infringed. They argue that hundreds of their compositions were included in the downloaded files, with the overall training allegedly affecting “tens of thousands” of protected musical works. The lawsuit requests a jury trial.
The use of BitTorrent introduces additional legal risks for Anthropic: the protocol not only downloads but also simultaneously distributes parts of the file to other users. This distinction is crucial for the plaintiffs, who argue that there was not only acquisition but also distribution of illegal copies, thus extending the dispute beyond the mere use of materials in training Claude.
In a prior federal case involving Anthropic, the court established a distinction between legally purchased books and pirated copies. The judge noted that purchasing physical copies, digitizing them, and using them for AI training could potentially fall under “fair use.” However, downloads from pirate libraries were characterized as “straightforward piracy on a colossal scale.” This distinction is once again at the forefront of the current case.
Anthropic has previously achieved a partial victory in a separate lawsuit regarding song lyrics filed in 2023, where the court declined to halt the training of Claude, and the company agreed to maintain protective filters on the model’s output. Since then, copyright disputes surrounding Anthropic have increasingly shifted focus toward the origins of training data—first concerning books and now extending to musical repertoire.
Read More: Ethereum Faces .3K Test Despite 12-Day ETF Inflow Streak.
