Publishers vs. Anthropic: Music Giants Move to Block AI ‘Fair Use’ Defense in $3B Suit

The legal definition of "fair use" in the age of generative AI is facing its most significant challenge yet. In a motion filed on Monday, March 23, 2026, in a San Jose federal court, a coalition of the world’s largest music publishers argued that Anthropic’s training methods do not meet the legal criteria for transformative use. The publishers—Universal Music Group, Concord, and ABKCO—assert that Anthropic didn't just "learn" from their lyrics; it systematically downloaded unauthorized copies of more than 20,517 copyrighted works from "shadow libraries" and torrent sites to build its Claude models. They are asking U.S. District Judge Eumi Lee to rule on these copyright infringements before the case even goes to trial.

The publishers' case centers on the argument that Claude’s output is not transformative, but substitutive. They provided evidence showing that Claude can be prompted to generate near-verbatim lyrics for iconic songs like Neil Diamond’s "Sweet Caroline" and The Rolling Stones’ "Wild Horses." By providing these lyrics to users on demand, the publishers argue that Claude acts as a direct competitor to licensed lyric websites and digital services, thereby "harming the potential market" for the original works—one of the four key pillars used to determine fair use under US law. They also allege that Anthropic willfully stripped "Copyright Management Information" (CMI) from the data, which is a separate violation of the Digital Millennium Copyright Act (DMCA).

The stakes are historically high, with the publishers seeking the maximum statutory penalty of $150,000 per infringed work, bringing the potential total to over $3.07 billion. This move follows a massive $1.5 billion settlement Anthropic paid to a group of authors in late 2025 for similar training data issues. Just days before this latest motion, music rights firm BMG also filed a separate lawsuit against Anthropic, citing 493 specific instances of lyric infringement involving artists like Bruno Mars and Ariana Grande. These parallel legal actions suggest a coordinated effort by the music industry to force AI companies into high-stakes licensing agreements.

Anthropic has previously defended its practices by comparing AI training to a human "reading" a book to learn how to write. However, the publishers argue that "reading" via mass-torrenting from pirate sites is not a protected activity. If the judge rules against Anthropic’s fair use defense, it could force a fundamental shift in how AI models are built, requiring developers to secure explicit licenses for every piece of data in their training sets. For Anthropic—recently valued at $380 billion—the outcome of this motion could define its financial future and the legal boundaries of the entire generative AI industry.