The legal battle between the music publishing industry and artificial intelligence companies has intensified, with three prominent publishers filing a new copyright lawsuit against AI firm Anthropic that could expose the company to more than $3 billion in damages. The case represents a significant escalation in an ongoing dispute over how copyrighted works are used to train large language models.
Universal Music Publishing Group, Concord Music Group, and ABKCO have brought the new lawsuit in California, accusing Anthropic of illegally downloading tens of thousands of copyrighted songs from piracy websites while building the training dataset for its AI chatbot, Claude. The publishers argue that these piracy-related claims fundamentally strengthen their copyright case, especially in light of recent court decisions that limit how far AI companies can rely on “fair use” defenses.
Why a New Lawsuit Was Filed
The publishers are not starting this fight from scratch. They first sued Anthropic in October 2023, alleging that the company infringed their copyrights by copying song lyrics without permission and by allowing its AI model to generate lyrics that closely resembled protected works.
However, that original case hit a procedural wall last year when a U.S. court refused to allow the publishers to amend their lawsuit to add new allegations related to piracy. At the time, the court determined that the proposed claims went beyond the scope of the existing litigation.
Rather than continuing under those constraints, the publishers chose a different path. The newly filed lawsuit focuses specifically on alleged piracy, allowing them to present evidence they say only became available later and to pursue significantly higher statutory damages.
Allegations of Large-Scale Copyright Infringement
At the heart of the new filing is the claim that Anthropic downloaded at least 20,517 copyrighted musical compositions from illegal sources. According to the publishers, these songs were obtained from piracy websites and incorporated into datasets used to train the Claude AI model.
Under U.S. copyright law, statutory damages can reach as high as $150,000 per infringement. If a court were to accept the publishers’ argument that each song represents a separate violation, the total damages could reach approximately $3.08 billion.
While that figure is eye-catching, it reflects the cumulative effect of thousands of alleged infringements rather than a single penalty. Even so, it would represent one of the largest copyright damage claims ever brought against an AI company.
Comparisons to the Book Authors’ Lawsuit
The lawsuit arrives against the backdrop of another high-profile legal fight involving Anthropic—this time with book authors. That separate case briefly raised the possibility of damages reaching into the trillions of dollars, largely because it was structured as a class action representing a vast number of authors.
By contrast, the music publishers’ case is much narrower. It involves only three companies and their respective catalogs, rather than the broader publishing industry. Still, legal observers note that the case could serve as a template for additional lawsuits if other music publishers decide to pursue similar claims.
Sony Music Publishing, the largest music publisher in the world, is not currently involved. If major rights holders were to file their own cases using the same legal strategy, the overall financial risk to AI companies could multiply quickly.
Dispute Over AI Safeguards
Beyond the issue of piracy, the publishers are also challenging Anthropic’s claims about the effectiveness of safeguards designed to prevent copyrighted material from appearing in AI-generated responses.
During the original lawsuit, Anthropic argued that Claude only reproduced copyrighted lyrics when prompted in very specific ways and that new technical restrictions had been put in place to prevent this from happening in the future.
The new lawsuit disputes those assertions. The publishers claim that Claude continues to produce infringing content and that its safeguards can be easily bypassed, allowing users to extract copyrighted lyrics and other protected material.
Fair Use and the Source of Training Data
Anthropic has consistently maintained that its AI training practices fall under the doctrine of fair use, which allows limited use of copyrighted material without permission under certain conditions. That argument received partial support in the book authors’ lawsuit, where a federal judge ruled that AI training could qualify as fair use—but only when the source material was lawfully obtained.
The ruling drew a clear distinction between books that Anthropic legally purchased and scanned, which were considered fair use, and millions of ebooks allegedly downloaded from piracy websites, which were not. That distinction significantly weakened the legal position of AI companies that rely on pirated content during training.
Claims of Withheld Evidence
In the new lawsuit, the music publishers allege that Anthropic failed to disclose key information during the discovery phase of the original case. According to the filing, Anthropic allegedly concealed the fact that it had used torrenting to download millions of pirated books, some of which included musical compositions owned by the publishers.
The publishers say they only learned about this activity after it was revealed during proceedings in the book authors’ case. When they attempted to update their original lawsuit to reflect this new information, Anthropic successfully blocked the amendment on procedural grounds.
Importantly, the publishers note that Anthropic did not deny engaging in the alleged torrenting activity, instead arguing that the new claims did not belong in the existing lawsuit. That argument ultimately led to the filing of the new case.
