Elvis Presley rights owners sue Anthropic over music used in AI training

The commercial rights owners of Elvis Presley have taken AI developer Anthropic to federal court in San Francisco, claiming the company copied 274 songs without licences to train its Claude language models.

Authentic Brands Group, which oversees Presley's intellectual property and catalogue rights, led the filing alongside partner entities. The lawsuit targets prominent recordings and compositions, including Can't Help Falling in Love, Don't Be Cruel, Return to Sender and All Shook Up. The plaintiffs seek statutory damages of up to $150,000 per song, which places the total claim at roughly $41 million.

Unlicensed data and direct outputs

The court filing focuses on two related problems. First, it alleges Anthropic scraped protected recordings and lyrics off the web to build its training datasets without seeking consent or offering payment to creators. Second, it alleges Claude continues to generate text that closely mimics original song structures, verses and hooks when users ask the chatbot to adapt existing music.

Lead attorney Richard Busch stated in the filing that the developer built its business model on unauthorised copying and argued that existing guardrails fail to stop Claude from delivering infringing outputs. Anthropic responded on Sunday that it rejects the allegations and plans to mount a full defence in court. In similar past disputes, the company has contended that training models on publicly available web content falls within fair use principles.

Unlike broader lawsuits organised by major music industry associations, this filing represents an individual estate taking direct legal action against a major frontier model developer.

What this means for Australian businesses

This case shows practical risks for Australian companies that rely on commercial generative AI tools for marketing, copy and internal workflows.

Australian copyright law does not include a broad fair use exemption like American law. Instead, Australian fair dealing exceptions remain narrow, covering research, study, news reporting, parody and critique. While American courts will decide whether scraping training data constitutes fair use, Australian businesses must look closely at how they use tool outputs.

When a model produces material that mirrors existing copyright works, your business can face legal responsibility if you publish or distribute that content commercially. To protect your operations:

• Check vendor terms of service to see what copyright indemnification applies when you use commercial tiers of tools like Claude or ChatGPT.
• Avoid prompting models to replicate existing commercial styles, specific branded songs or creative work that belongs to another entity.
• Implement internal review checks so staff do not paste generated marketing copy, scripts or campaign collateral directly onto public channels without verifying its originality.
• Ensure company style guides remind team members that AI assistance does not replace proper licensing for creative assets.