US music publishers suing Anthropic make their case against AI 'fair use' - Reuters
BMG Rights Management has filed a copyright lawsuit against Anthropic in California federal court, alleging the AI company used pirated copies of songs by the Rolling Stones, Bruno Mars, Ariana Grande, and other artists to train its Claude chatbot, and that Chief Executive Dario Amodei personally approved the practice of sourcing training data from unauthorized libraries rather than paying licensing fees.
The lawsuit, filed March 17 in the Northern District of California, cites 493 compositions that BMG claims were copied and reproduced without permission. According to the complaint, Anthropic scraped lyrics from the internet and downloaded sheet music and songbooks via shadow libraries including Library Genesis, a platform widely known for distributing copyrighted material without authorization. BMG further alleges Anthropic removed copyright management information from these materials before incorporating them into its training data. A cease-and-desist letter BMG issued in December 2025 went unanswered, according to the filing.
The case is the latest in a wave of copyright infringement suits against AI companies and is closely related to an ongoing lawsuit brought by Universal Music Group, Concord, and ABKCO, which together allege infringement across more than 20,000 songs and seek more than $3 billion in damages. BMG is asking the court to assign its case to the same judge. Anthropic settled a separate class action from authors for $1.5 billion in September 2025, a figure that provides a benchmark for the potential financial exposure in music cases.
BMG said in a statement that Anthropic practices stand in direct opposition to responsible AI development, and that the company is knowingly engaging in widespread and systematic infringement. Statutory damages under U.S. copyright law can reach $150,000 per work if infringement is found to be willful, which in this case would substantially multiply the potential damages beyond the compensatory figure.
The core legal question is whether Anthropic fair use defense survives the specific allegation that it sourced material from pirated databases rather than licensed sources. Courts have not yet definitively ruled on whether AI training on copyrighted material constitutes fair use, but the shadow library sourcing allegation adds a dimension that the broader AI copyright cases have not yet squarely faced. If a court finds that using pirated sources specifically undermines the fair use argument, it would create a distinct legal risk for every AI company that has used unauthorized training data.