ANTHROPIC IS FACING A LAWSUIT OVER LYRICS FROM 853 SONGS
From Madonna to Joni Mitchell, the music industry is challenging the way artificial intelligence learns from copyrighted work — and the outcome could change how AI companies build their businesses.
1. A NEW COPYRIGHT BATTLE FOR THE COMPANY BEHIND CLAUDE
On October 21, 2026, music publisher Reservoir Media filed a lawsuit against Anthropic, the company behind the AI assistant Claude.
The allegation is serious: Reservoir claims that Anthropic used the lyrics of 853 copyrighted musical compositions to train its AI models without permission.
The catalogue includes works associated with major songwriters and internationally recognised music. Reservoir argues that the company benefited from creative work that it did not have the right to copy.
Anthropic disputes the allegations and has said it intends to defend itself vigorously.
At this stage, these are claims in a lawsuit, not a finding that Anthropic has infringed the copyrights. A court will need to assess the evidence and the legal arguments.
But the case raises a much bigger question: Can an AI company learn from copyrighted music without paying the people who created it?
2. WHY SONG LYRICS ARE VALUABLE TO AI COMPANIES
An AI assistant such as Claude learns patterns from large collections of text. By processing language, it can learn how people communicate, explain ideas, write stories and respond to questions.
Song lyrics contain a distinctive form of language. They use rhythm, repetition, rhyme, imagery, emotional expression and storytelling to communicate ideas in relatively few words.
These characteristics can make lyrics useful examples of creative writing.
However, learning from a song and reproducing its lyrics are not necessarily the same thing.
AI companies argue that training can transform existing material into a system capable of performing new tasks. Copyright owners counter that copying protected works without permission can violate their rights, particularly when a model can reproduce or imitate the material in ways that compete with licensed products.
Reservoir's lawsuit focuses on both the alleged use of lyrics during training and the company's claims about how Claude can generate song-related text.
The distinction matters because the legal questions surrounding training data can differ from those surrounding the content an AI system produces.
3. THE 853 SONGS AT THE CENTRE OF THE DISPUTE
Reservoir is an independent music publisher with a catalogue covering songs from different eras and genres.
Its complaint identifies 853 compositions whose lyrics it alleges were used without authorisation.
The dispute is part of a wider confrontation between the music industry and AI companies. Publishers want to protect the commercial value of the songs they control, while AI developers want access to enough material to build systems capable of understanding and generating language.
The financial stakes can be substantial. Music publishers earn revenue through licensing arrangements for uses such as recordings, performances, reproductions and other authorised uses of musical compositions.
If AI companies can use song lyrics without paying for licences, publishers fear that an important category of commercial rights could lose value.
If every use of copyrighted material during AI training instead required individual permission, developers could face substantial costs and complicated negotiations involving enormous collections of works.
Finding a workable legal boundary is therefore important to both sides.
4. THE DAMAGES COULD REACH $151.4 MILLION
Reservoir is seeking damages that could reach approximately $151.4 million under its legal claims.
That figure is a maximum demand based on the company's allegations and the damages potentially available under applicable law. It is not a judgment, a confirmed loss or money Anthropic has been ordered to pay.
Copyright law in the United States can permit statutory damages of up to $150,000 for each work in cases of willful infringement, subject to the legal requirements and circumstances of the case.
Reservoir also raises claims under the Digital Millennium Copyright Act, which addresses certain forms of circumvention of access controls and removal of copyright-management information.
The publisher's calculation depends on which claims succeed and which remedies the court considers available. The final amount, if any, could be substantially different.
For Anthropic, however, the potential exposure is about more than one lawsuit. Multiple disputes involving copyrighted material can create legal expenses, uncertainty and pressure to negotiate licences or change how training data is obtained.
5. WHY THIS IS NOT AN ISOLATED CASE
Anthropic has already faced legal challenges from music publishers over the use of copyrighted lyrics.
In January 2026, Universal Music Group, Concord and ABKCO brought another major lawsuit alleging that Anthropic had copied protected musical works. That case covered a much larger catalogue and raised claims involving alleged piracy and the use of lyrics to train Claude.
Other publishers, including Sony Music Publishing and Warner Chappell Music, have also pursued legal action over the alleged use of copyrighted songs in AI training.
The growing number of disputes shows that the industry is not relying on a single legal strategy. Different publishers are testing different claims, evidence and arguments.
That creates uncertainty for AI companies. Even if a court rules favourably on one aspect of training, other issues — such as how the data was acquired or whether a model reproduces protected lyrics — may still require separate consideration.
6. THE BIG QUESTION: DOES AI TRAINING QUALIFY AS FAIR USE?
One of the most important legal arguments concerns fair use, a US copyright doctrine that can permit certain uses of copyrighted material without permission.
AI companies argue that training models on large collections of text can be transformative because the objective is to learn patterns and develop new capabilities, rather than simply republish the original material.
Publishers argue that copying their work without permission can undermine the market for licensing, particularly if AI products reproduce lyrics or provide alternatives to services that already pay for access.
Courts must consider several factors, including the purpose of the use, the nature of the copyrighted work, how much was copied and the effect on the potential market.
The answer is not necessarily the same for every situation.
A court might distinguish between using a work to train a model, obtaining it from an unauthorised source and generating a response that reproduces protected expression. Those activities raise related but different questions.
The outcome could influence how future AI products are developed, trained and commercialised.
7. WHAT THIS MEANS FOR THE FUTURE OF AI
The dispute highlights a business problem that AI companies can no longer ignore: high-quality training data has economic value, and the people who create it increasingly want a say in how it is used.
One possible outcome is a larger market for licensing agreements. Publishers could negotiate payments that allow AI developers to use their catalogues under defined conditions.
Another possibility is that companies invest more heavily in authorised datasets, content partnerships and technical safeguards designed to reduce the risk of reproducing protected material.
These approaches could increase development costs, but they might also give AI companies greater legal certainty.
For musicians and songwriters, the question is whether the growth of AI will create new income opportunities or weaken existing sources of revenue.
For businesses, the lesson is broader: technological innovation does not automatically remove the legal and commercial rights attached to the material that makes innovation possible.
MAACAT PERSPECTIVE
Anthropic's lawsuit over 853 songs is part of a much larger debate about who should benefit when artificial intelligence learns from human creativity.
AI companies need data to build useful products. Musicians and publishers need ways to protect the work that supports their livelihoods. Both interests have real economic value, and the legal system is still working through where the boundaries should lie.
The companies that navigate this issue successfully may be those that build strong technology while establishing sustainable relationships with the creators whose work helps make it possible.
The future of AI will not depend only on how much a model can learn. It will also depend on who has the right to teach it.
