Sony Music and Warner Sue Anthropic for Up to $150,000 Per Copyrighted Work

Sony Music and Warner Chappell have filed a federal copyright lawsuit against Anthropic, demanding up to $150,000 per infringed work across "tens of thousands" of songs — a potential damages figure that could climb into the billions and set a pricing floor for what AI training on creative work actually costs.

Key takeaways

• Sony Music and Warner Chappell sued Anthropic in the US District Court for the Northern District of California on August 29, 2026.
• The labels are seeking statutory damages of up to $150,000 per copyrighted work, plus up to $25,000 for each instance where copyright-management information was stripped.
• The suit covers "tens of thousands" of works, making the total potential liability among the largest ever sought in an AI copyright case.
• The complaint uses the phrase "brazen campaign" of intellectual property theft, signaling the labels intend to pursue the most aggressive framing available under copyright law.
• This is a direct consequence of AI models being trained on unlicensed text corpora — a practice that affects every major foundation model, including those powering image and audio generation tools.

The damages math that should alarm every AI lab

According to The Verge, the complaint was filed in the Northern District of California — the same venue that has handled most of the high-profile AI copyright cases to date. The $150,000-per-work figure is the statutory maximum under US copyright law for willful infringement; the labels are not asking for actual damages, which are harder to quantify. They're going straight for the ceiling.

The additional $25,000 per instance of stripped copyright-management information — metadata that identifies who owns a work — suggests the complaint will argue that Anthropic didn't just train on protected material, but actively removed the fingerprints that would have made the infringement traceable. That's a harder allegation to defend against than simple ingestion.

Why this case is structurally different from earlier music-AI suits

When Stability AI faced copyright challenges in 2023 over Stable Diffusion's training data, the core dispute was whether image generation constituted derivative infringement. The Sony-Warner complaint against Anthropic homes in on something more specific: direct reproduction of lyrics and musical works by a large language model, framed explicitly as "piracy" rather than transformation. That framing matters legally because it bypasses fair-use arguments that have given AI companies room to maneuver.

For creators who use AI tools — whether for generating images, building characters, or producing music — the lawsuit is a signal that the licensing question that has hovered over every major foundation model is now being litigated at maximum statutory scale. The outcome will determine whether AI companies need blanket licensing deals before training, or whether they can continue to argue that training is transformative use and settle afterward.

Anthropically has already had a turbulent legal year: a federal court blocked a separate government attempt to blacklist the company earlier in 2026. That ruling was about government overreach; this lawsuit is about private property rights, and the two legal theaters are entirely separate.

What the labels want — and what it forces Anthropic to weigh

The labels' ask is not a licensing deal. It's punitive damages designed to make the cost of unlicensed training retroactively prohibitive. If even a fraction of the "tens of thousands" of works cited in the complaint survive summary judgment at the $150,000 rate, the liability would dwarf anything Anthropic has raised in venture funding.

That pressure is relevant beyond Anthropic. Every AI company training on web-scraped text — including those whose models underpin the tools in Charmloop's catalog — faces the same structural exposure. A ruling or settlement that establishes a per-work damages rate would immediately reprice the cost of building foundation models, likely accelerating the shift toward licensed datasets or synthetic training data.

The case now moves to the Northern District of California. The first major procedural test — whether the court certifies the scope of the complaint or narrows it — will determine how much leverage the labels retain at the negotiating table. That hearing, not any eventual trial, is the near-term date worth watching.