# Sony and Warner Chappell are taking the fight to Anthropic over > Source: > Published: 2026-08-29 19:15:00+00:00 # Sony and Warner Chappell are taking the fight to Anthropic over [Claude](/en/tags/claude/)models without permission or compensation. This isn't just a minor dispute over a few snippets of lyrics; it’s a fundamental challenge to how generative AI models are built and whether the "fair use" defense can actually hold up when it involves massive, high-value creative catalogs. If you follow the ongoing litigation involving OpenAI and various media giants, you’ll notice a pattern emerging. The core of the argument from the music industry is that AI companies are essentially "ingesting" the lifeblood of the creative economy to create tools that might eventually compete with the very creators they learned from. For Anthropic, the stakes are high because their focus on "Constitutional AI" and safety is supposed to set them apart, but safety doesn't necessarily equate to legal compliance regarding intellectual property. ## The technical friction point The friction lies in the data ingestion phase of the AI workflow. When an LLM is being trained, it scrapes massive datasets to learn patterns, structures, and nuances. In the case of music, this includes: **Lyrical structures:** The rhythm, rhyme schemes, and semantic meaning of songs.**Melodic patterns:** While models like Claude are text-based, the descriptions or transcriptions of musicality can be part of the training set.**Copyrighted compositions:** The specific expression of ideas that are legally protected. The publishers argue that Anthropic's models can generate content that mimics the style or substance of protected works, which they claim is a direct result of unauthorized training. ## Why this matters for prompt engineering and AI development For those of us working deep in prompt engineering or building LLM agents, this legal shift could change how we approach model capabilities. If the courts rule in favor of the publishers, we might see: 1. **Stricter data provenance:** Developers will have to provide much more transparent documentation on what went into their training sets. 2. **Licensed datasets becoming the standard:** Instead of massive, unregulated web scrapes, we might move toward highly curated, paid-for datasets. This could lead to a "walled garden" effect where only the wealthiest AI companies can afford the best training data. 3. **Changes in model output:** We might see more aggressive "guardrails" implemented within the model's constitutional layer to prevent it from outputting anything that even slightly resembles copyrighted lyrical patterns. This isn't just about music; it's a test case for the entire generative AI industry. If Anthropic loses, the cost of training a frontier model could skyrocket, potentially slowing down the deployment of new, highly capable models. On the other hand, if they win, it sets a precedent that "learning" from public data is fundamentally different from "copying" it. This distinction is exactly what the legal teams on both sides are fighting to define. [Music publishers are taking Anthropic to court over alleged 48m ago](/en/news/8160/) [OpenAI and Anthropic are driving San Francisco's rent through 16h ago](/en/news/8074/) [The NSA is pushing for a backdoor into every AI model in 22h ago](/en/news/8050/) [A judge just stepped in to stop the Pentagon from blacklisting 1d ago](/en/news/7999/) [Meta might drop $10B to secure Anthropic's expertise 1d ago](/en/news/7949/) [My brain is rotting because of Claude Code and the pressure to 2d ago](/en/news/7923/) [Next Music publishers are taking Anthropic to court over alleged →](/en/news/8160/)