Music Publishers File for Partial Summary Judgment Against Anthropic Over AI Training Data
Bot Mutiny |
A group of major music publishers, including Concord and Universal, submitted a separate statement of facts claiming Anthropic scraped and stored thousands of copies of copyrighted lyrics to train its Claude AI models.
Concord Music Group, Inc. and a coalition of the world's largest music publishers have moved for partial summary judgment in their copyright infringement lawsuit against Anthropic PBC. The plaintiffs filed a separate statement of undisputed facts on September 22, 2026, in the U.S. District Court for the Northern District of California. The filing, first reported by CourtListener, outlines the publishers' evidence regarding how Anthropic built its Claude AI models. The plaintiffs include Abkco Music Inc, Capitol CMG, Inc., Polygram Publishing, Inc., Songs of Universal Inc., Universal Music - MGB NA LLC, Universal Music - Z Tunes LLC, and Universal Music Corp. ## Training Data and Web Scraping The publishers allege Anthropic unlawfully reproduced 499 specific musical works to serve as training input for Claude. According to the filing, Anthropic admitted in discovery that at least one Claude model was trained on a dataset containing the lyrics to at least 100 of the works listed in the suit. The publishers' statement claims Anthropic doesn't deny that the lyrics to these works are included in Claude's training data. The filing asserts that Anthropic assembled this collection by copying and downloading lyrics from the internet and third-party datasets. Anthropic uses automated tools, including web crawlers, to scrape text from the internet at a large scale. The publishers argue this process involves downloading protected lyrics directly onto Anthropic's servers. ## Scale of the Infringement An inspection of the dataset or "corpus" used to train Claude identified specific copies of the publishers' lyrics. The publishers state their inspection found that Anthropic's training data contained lyrics for each of the 499 works identified in the litigation. The volume of reproduction cited in the document is significant. For 354 of the works, the publishers claim Anthropic's training data contained over a certain number of copies of each. The specific counts for these copies were redacted in the public version of the filing. The publishers contend these lyrics were registered with the Copyright Office within five years of their initial publication. They also verified the accuracy of the lyrics by comparing them to data from LyricFind.com, a lyric aggregator. ## Legal Arguments The motion for partial summary judgment focuses on a prima facie case of direct infringement. The publishers argue there's no dispute that they own or exclusively control the rights to the lyrics in these 499 works. The case, Concord Music Group, Inc. v. Anthropic PBC, is currently before Judge Eumi K. Lee and Magistrate Judge Susan van Keulen in the San Jose Division. The filing follows a joint administrative motion to file certain documents under seal related to the summary judgment and Daubert motions. Anthropic's response to these specific factual claims remains part of the ongoing litigation. The publishers' statement relies on deposition testimony from Anthropic employees and the company's amended responses to requests for admission.