Anthropic Seeks Summary Judgment on Fair Use in AI Music-Lyrics Dispute
Universal Music Group, Concord and ABKCO have asked a US federal court to determine that Anthropic's alleged use of copyrighted song lyrics to train its Claude chatbot is not protected by the fair-use doctrine. Anthropic has advanced the opposite position and also sought summary judgment on the dispute.
US District Judge Eumi Lee heard arguments from the parties in San Jose on 21 October. The judge questioned both sides on the nature of the training use, the asserted transformation in Claude's outputs and the potential effect of AI-generated lyrics on the market for protected songs. She did not decide the case or issue a ruling at the hearing.
The publishers filed the dispute in 2023 and allege that Anthropic used lyrics from hundreds of songs, including works associated with Beyoncé, the Rolling Stones and the Beach Boys, to train Claude. Through counsel Sonal Mehta of WilmerHale, Anthropic argued that the use is transformative because the training results in the production of new content rather than the simple reproduction of lyrics.
Anthropic also maintained that AI training does not reduce demand for songs or their lyrics. The publishers, represented by Matt Oppenheim of Oppenheim + Zebrak, contended that the same training enables Claude to create synthetic lyrics that compete with protected works. The publishers placed reliance on nearly 120,000 requests made to Claude over six months for the generation of synthetic lyrics and warned that AI systems could flood the market with competing material.
Is using copyrighted song lyrics to train an AI model fair use?
The fair-use question remains undecided. Anthropic argues that training produces transformative output and does not substitute for the original songs, while the publishers say Claude can generate competing lyrics and thereby harm the market for those works. Judge Eumi Lee heard the parties’ summary-judgment arguments on 21 October, but the US District Court for the Northern District of California has not yet determined whether the challenged use qualifies as fair use.
Competing US rulings on AI training
Other judges in the same judicial district have reached differing conclusions on AI training. The evidence presented in the dispute records two distinct approaches adopted during 2025, although neither ruling determines the outcome of the publishers’ claim against Anthropic.
Authority or proceeding | Position on AI training and copyright |
|---|---|
William Alsup, US District Judge | Ruled that training on pirated copies is not fair use, while the use of lawfully acquired books for AI training may qualify as fair use. |
Vince Chhabria, US District Judge | Ruled in a comparable dispute that AI training does not, in many cases, constitute fair use. |
UMG, Concord and ABKCO v Anthropic | Anthropic contends that its use of copyrighted song lyrics for training Claude is fair use. The publishers dispute that position. Judge Eumi Lee has heard summary-judgment arguments but has not ruled. |
The underlying fair-use issue therefore remains subject to further judicial analysis. Anthropic also faces separate proceedings brought by Sony Music Publishing and Round Hill Music involving song lyrics.
India’s proposed licensing framework is not yet law
The position in India differs from the general fair-use doctrine applied in the United States. The Copyright Act, 1957 contains specific fair-dealing exceptions under Section 52 rather than a general fair-use defence.
On 8 December 2025, the Department for Promotion of Industry and Internal Trade issued Part I of a working paper on generative AI and copyright. The proposal described a mandatory blanket licence under the formulation “one nation, one licence, one payment”. Rights holders would not be permitted to withhold lawfully accessed works from participation. Royalties would instead be determined by a government-appointed committee when an AI system is commercialised.
The document was opened for public comments and stated that a second part would follow. The evidence does not identify the publication of Part II, and the proposal has not become law.
What is the status of AI-copyright proceedings in India?
Indian courts have not yet determined whether copyrighted works may lawfully be used to train AI models. The ANI v OpenAI dispute concerning ChatGPT training remains pending, with the Indian Music Industry, T-Series and Saregama appearing as intervenors. The status of that proceeding means that no Indian court ruling presently answers the legality of using protected lyrics or recordings for AI training.
Licensing and safeguards remain unresolved
Questions remain over whether Anthropic has obtained licences for Indian song lyrics or music catalogues and whether Claude contains safeguards against reproducing lyrics in Indian languages. The publishers have also been asked how the figure of nearly 120,000 synthetic-lyric requests was measured, whether it forms part of the court record and whether the publishers or their Indian affiliates intend to initiate similar proceedings in India.
The Indian Music Industry has been asked whether AI developers have approached it or its members for training licences and on what terms. The proposed licensing framework also raises questions about royalty rates for AI outputs that compete with original songs. These issues have not been finally resolved by the US court or determined in the pending Indian proceedings.