Sony and Warner sue Anthropic alleging massive music piracy

Generative AI copyright litigation has evolved through distinct waves corresponding to media type and ingestion methodology. Early cases focused on text corpora and web scraping, establishing initial precedents around fair use and database assembly. Subsequent filings expanded to visual art and code repositories, often centering on output similarity and opt-out mechanisms. The current wave targets audio and musical composition rights, introducing novel allegations regarding reinforcement learning workflows and synthetic data pipelines as vectors for infringement. Prior settlements in the text domain have created a bifurcated landscape where resolved liabilities for one modality do not necessarily immunize companies against parallel claims in adjacent creative fields. Legal theories have simultaneously shifted from pure reproduction claims toward allegations of systematic circumvention of technical protection measures and intentional inducement of infringement through internal corporate directives. This pattern reflects an iterative adaptation of copyright enforcement strategies as plaintiffs refine their understanding of model architecture and training supply chains.

On August 29, 2026, units of Sony Music Publishing and Warner Chappell filed a federal lawsuit in the Northern District of California against Anthropic, its CEO Dario Amodei, and co-founder Benjamin Mann. The complaint alleges that the defendants engaged in what the plaintiffs describe as 'one of the largest and most blatant ongoing thefts of intellectual property in history' to train their Claude artificial intelligence models. According to the filing, Anthropic conducted a campaign of illegally torrenting, scraping, and downloading copyrighted musical works on a massive scale to develop and profit from its AI systems. The plaintiffs are seeking statutory damages of up to $150,000 per infringed work, additional penalties of up to $25,000 for each instance of stripped copyright management information, court-ordered disclosure of training data and collection methods, and the destruction of infringing copies.

The lawsuit makes specific factual allegations regarding internal decision-making at Anthropic. According to the complaint, co-founder Benjamin Mann discussed acquiring data from LibGen, a shadow library known for hosting pirated content, with CEO Dario Amodei, who allegedly approved the torrenting of this material. The filing asserts that an earlier court had already found Mann knew the books in question were pirated. Furthermore, the plaintiffs allege that this pirated material from LibGen and another source identified as PiLiMi reached commercial versions of Claude indirectly. The mechanism described involves synthetic data generation and reinforcement feedback derived from other Anthropic models that were originally trained on the unauthorized datasets, effectively laundering the infringing content into production systems.

A central technical allegation concerns the fine-tuning process used to align Claude’s outputs. The complaint claims that Anthropic stripped copyright-identifying information from training materials and specifically rewarded accurate reproduction of copyrighted lyrics during model refinement. According to the filing, human reviewers were instructed to select accurate lyric reproductions over inaccurate ones during the fine-tuning phase, thereby incentivizing verbatim recall of protected musical compositions. This practice, the plaintiffs argue, demonstrates willful infringement rather than incidental ingestion, distinguishing the case from broader fair use debates about general knowledge acquisition.

This legal action arrives less than a year after Anthropic agreed to a $1.5 billion settlement in September 2025 to resolve separate copyright claims brought by authors and publishers regarding text-based training data. That agreement was reported as the largest copyright settlement in U.S. history. However, the new music industry suit tests whether such prior resolutions provide any protective umbrella for different media types or if each content vertical represents a distinct liability frontier. As of the filing date, Anthropic has not publicly responded to the specific allegations contained in the newly submitted complaint. The case is positioned as an opening salvo in what legal observers anticipate will be a protracted conflict over intellectual property rights in the generative AI era, specifically testing how courts treat allegations of direct piracy versus transformative use in the context of large language model training.

Sony Music and Warner Chappell are pursuing statutory damages up to $150,000 per work plus $25,000 per CMI violation across tens of thousands of compositions, creating potential liability exceeding prior settlements. Anthropic risks compelled disclosure of proprietary training data, RLHF protocols, and executive communications, alongside possible destruction orders for model weights. The outcome may determine whether AI firms can operate with media-specific licensing or face consolidated cross-modal liability. Individual executive liability introduces personal risk beyond corporate exposure. The music industry seeks to establish precedent that reinforcement learning incentives constitute direct infringement, potentially reshaping alignment practices industry-wide.