Sony Music Publishing and Warner Chappell sue Anthropic for using tens of thousands of copyrighted songs to train Claude, alleging brazen intellectual property theft.

The legal battle between the generative AI industry and the guardians of intellectual property has entered a high-stakes new chapter. Sony Music Publishing and Warner Chappell Music have launched a multibillion-dollar federal lawsuit against Anthropic, the startup behind the Claude chatbot. The complaint, filed in the U.S. Northern District of California, describes a systematic effort to acquire copyrighted musical compositions without authorization, characterizing the AI company's actions as one of the most blatant thefts of intellectual property in history.

A brazen campaign of digital piracy

At the heart of the lawsuit is the allegation that Anthropic engaged in a widespread campaign of illegally torrenting, scraping, and downloading copyrighted works. The plaintiffs claim that Anthropic did not merely stumble upon public data but actively sought out protected material to fuel the training of its Claude models. This process allegedly involved the use of pirate websites, specifically naming Library Genesis and the Pirate Library Mirror as sources for lyrics and sheet music.

The scale of the alleged infringement is vast. The publishers assert that tens of thousands of songs were misappropriated. Among the cited works are global smash hits that define various eras of popular music, such as Mariah Carey's All I Want for Christmas is You, Survivor's Eye of the Tiger, and the classic collaboration between Marvin Gaye and Tammi Terrell, Ain’t No Mountain High Enough. The lawsuit further alleges that the AI model can reproduce these copyrighted lyrics in response to user prompts, proving that the protected content was not just analyzed but absorbed into the model's output capabilities.

From pirate libraries to licensed scrapers

The legal filing suggests a multi-pronged approach to data acquisition. While the use of torrents and pirate mirrors highlights a disregard for copyright law, the publishers also point to the scraping of legal, licensed lyric websites. Variety reports that the suit specifically mentions Musixmatch and LyricFind, as well as the use of archive datasets like Common Crawl.

This distinction is critical for the court. By targeting both pirate repositories and licensed platforms, the plaintiffs argue that Anthropic bypassed the established economic ecosystem of the music industry. The lawsuit names not only the corporation but also CEO Dario Amodei and co-founder Benjamin Mann as individual defendants, suggesting a level of leadership awareness and direction in these data acquisition strategies.

The ghost of the Bartz settlement

This is not the first time Anthropic has faced the consequences of its training methods. The current lawsuit draws a direct line to a previous legal conflict with U.S. authors, known as Bartz v. Anthropic. In that instance, the company eventually settled for .5 billion. The music publishers are leveraging the facts from that case, noting that it was revealed Benjamin Mann and Anthropic had downloaded at least 7 million copies of books from pirate websites.

Defendants can no longer hide their extraordinary theft, and their mass infringement is now well-documented.

The precedent set in the Bartz case is a double-edged sword for the AI firm. While the judge in that matter ruled that training models on copyrighted content could potentially be permissible, the court explicitly stated that obtaining such content through piracy was not protected. This specific nuance is where Sony and Warner are focusing their attack, arguing that the method of acquisition—torrenting and scraping—negates any fair use defense.

Anthropic's defense and the fair use argument

Anthropic has not remained silent in the face of these allegations. A company spokesperson characterized the lawsuit as a recycling of claims already being litigated in other courts. The core of their defense rests on the concept of transformative fair use. They argue that the process of training a large language model (LLM) creates something entirely new and different from the original source material, thereby not infringing on the copyright of the individual songs.

However, the music industry is increasingly less patient with the fair use argument. The publishers are seeking statutory damages of up to 0,000 per infringing work. Given that the suit mentions tens of thousands of songs, the financial exposure for Anthropic could reach staggering heights, potentially dwarfing the previous author settlement.

A pattern of industry-wide litigation

The clash between Anthropic and the major publishers is part of a broader trend of reckoning within the music business. The industry has oscillated between aggressive litigation and strategic licensing deals as it attempts to navigate the rise of generative AI. Fortune notes that this case follows similar actions brought by BMG, Round Hill Music, and a group including Universal Music Group and Concord Music Group.

The landscape is fragmented. While some companies, like Warner Music Group, have settled cases against AI music generators like Suno and Udio, others remain in active litigation. Universal Music Group has taken a hybrid approach, striking a deal with Udio while continuing to pursue Suno in court. This suggests that the music industry is using lawsuits as a leverage tool to force AI companies toward licensing agreements that ensure songwriters and composers are compensated for their intellectual property.

Global business implications for AI and IP

For entrepreneurs and business leaders in the USA, UK, and global markets, this lawsuit serves as a critical warning regarding the procurement of training data. The era of the wild west of scraping is rapidly closing. In the United States, the focus remains on the interpretation of fair use, but the Bartz v. Anthropic precedent suggests that the source of the data is just as important as the use of the data. Using pirated datasets is increasingly viewed as an indefensible legal liability.

In the UK and other global jurisdictions, the trend is moving toward stricter transparency requirements. Companies deploying AI models must now consider the provenance of their training sets to avoid multibillion-dollar liabilities. For businesses building on top of LLMs, this creates a secondary risk: if the underlying model is found to be trained on stolen IP, the legal fallout could impact the stability and availability of the tools they rely on. The shift toward licensed data ecosystems is no longer optional; it is a necessity for long-term corporate viability in the AI sector.

FAQ

---

Sources: Theguardian, Fortune, Finance ·