Major Record Labels Sue Anthropic and Founders Personally Over Lyric Training Data

The legal confrontation between the creative entertainment industry and generative artificial intelligence labs reached a high-stakes turning point as two of the world's most dominant music publishing companies took aim not just at a corporate balance sheet, but directly at executive leadership. In a comprehensive 48-page federal lawsuit filed in the U.S. District Court for the Northern District of California, Sony Music Publishing and Warner Chappell Music accused AI research company Anthropic of massive copyright infringement, claiming the company unlawfully ingested vast catalogs of pirated song lyrics and sheet music to train its Claude family of large language models. Critically, the complaint takes the rare and aggressive step of naming Anthropic co-founder and Chief Executive Officer Dario Amodei alongside co-founder Benjamin Mann as individual defendants, demanding maximum statutory damages of up to $150,000 per infringed work—a sum that, given the tens of thousands of musical compositions cited, theoretically exposes the AI company to billions of dollars in liability.

Background

To build cutting-edge large language models (LLMs) such as Claude or OpenAI's GPT series, engineers train neural networks on monumental volumes of text scraped from across the internet, public datasets, digitized books, and web archives. This training allows models to grasp grammar, factual knowledge, stylistic nuance, and creative cadence. However, this process has sparked an existential conflict over intellectual property rights.

Tech developers contend that scraping publicly available text to teach algorithms constitutes "fair use"—a legal doctrine that allows limited use of copyrighted material without permission if the new work is deemed transformative and does not directly displace the market for the original. Content creators, publishers, and media conglomerates fiercely counter that ingestive copying constitutes blatant theft without licensing agreements or royalties, especially when the resulting models can generate synthetic text, prose, and lyrics that directly compete with human creators.

The music industry has been on the front line of this dispute. Universal Music Publishing Group, Concord Music Group, and ABKCO initially filed suit against Anthropic in late 2023 over lyric ingestion. While Anthropic previously navigated a landmark $1.5 billion settlement to resolve class-action claims brought by book authors over training sets like Books3 and pirate library repositories, music publishers maintain that the company has continued to leverage protected creative catalogs without commercial licenses.

What Happened

According to the newly filed 48-page complaint reported by Music Business Worldwide and Business Insider, Sony Music Publishing and Warner Chappell allege that Anthropic systematically harvested protected musical compositions from pirate repositories—such as Library Genesis and Pirate Library Mirror—as well as web crawls and unauthorized digital databases. The lawsuit points to thousands of iconic compositions spanning decades of pop culture, including Mariah Carey’s "All I Want for Christmas Is You," Survivor's "Eye of the Tiger," Bon Jovi's "Livin' on a Prayer," Leonard Cohen's "Hallelujah," and Taylor Swift's "Paper Rings."

The plaintiffs characterize the activity as "one of the largest and most blatant ongoing thefts of intellectual property in history." Beyond the general ingestion, the lawsuit stresses that Claude reproduces copyrighted lyrics verbatim upon user prompting. While Anthropic has implemented guardrails and system filters designed to curb direct copyright outputs, the complaint contends that these protections are easily circumvented through basic re-prompting.

Crucially, the legal action pierces the corporate veil by naming Dario Amodei and Benjamin Mann personally. The complaint cites discovery materials and planning documents unsealed in related litigation, alleging that Mann directly downloaded pirated databases to circumvent the cost of purchasing them, and referencing internal notes suggesting an effort to conceal the scraping and scanning operations. Alongside maximum statutory damages of up to $150,000 for each willfully infringed composition and $25,000 per violation for altering copyright management information, the music publishers are demanding a jury trial, an exhaustive audit of Claude's training data, and the court-ordered destruction of all infringing copies.

Anthropic has pushed back firmly against the claims. A spokesperson stated that the lawsuit recycles arguments already being litigated by the same counsel, reiterating that model training represents fair use and affirming that the lab will vigorously defend itself in court.

Why It Matters

The decision by Sony and Warner Chappell to name Amodei and Mann personally marks a seismic tactical escalation. In typical corporate copyright battles, executive leadership remains insulated behind limited liability structures. By targeting founders individually under theories of direct and contributory infringement, the music publishing giants are attempting to establish individual financial and personal accountability for high-level data curation decisions made during model development.

Furthermore, the case tests the boundaries of the fair use doctrine when applied to unauthorized and pirated data sources. While AI companies have argued that model training transforms raw text into mathematical representations and abstract concepts, courts may view the use of illicit torrent repositories differently than standard web crawling. In Europe, the Regional Court of Munich recently ruled against OpenAI in a lyric reproduction case, finding that memorizing lyrics is an act of reproduction beyond statutory text-and-data-mining exemptions, especially where access is unlawful. If American federal judges similarly determine that training on pirated repositories precludes fair use protections, AI labs could face existential liabilities across their entire legacy data pipelines.

Economically, the multi-billion-dollar statutory ceiling sought by plaintiffs illustrates that copyright litigation is no longer viewed as an afterthought that AI startups can absorb as a minor operating cost. Instead, it threatens the legal foundation of foundational models that were built prior to formal licensing frameworks.

What's Next

In the immediate term, Anthropic is expected to file motions to dismiss, particularly targeting the personal claims against Amodei and Mann while arguing that the claims duplicate ongoing litigation brought by other music groups. How the court handles the personal liability aspect will be scrutinized closely by startup founders, venture capitalists, and board members across Silicon Valley.

Moving forward, the pressure exerted by these simultaneous lawsuits will likely accelerate the commercial push toward bilateral data licensing deals. Just as digital streaming services were eventually forced into structured royalty-sharing agreements with record labels two decades ago, AI developers may find that negotiating direct licensing pacts is the only viable path to avoid catastrophic court judgments and injunctions.

In the meantime, the trial proceedings in Northern California will serve as a bellwether for the entire tech sector. Whether through a sweeping courtroom precedent or another massive structured settlement, the outcome of this dispute will define who owns—and who gets paid for—the culture feeding the world’s most advanced artificial intelligence.