Skip to main content
AI doesn’t set its own limits. We do.
AI risk2 min read

Elvis Song Owners Sue Anthropic Over Training Data

A copyright action over 274 compositions highlights a simpler buyer question: can you prove where your AI training and reference data came from and that use was licensed?

Short answer

Copyright disputes over AI training are unsettled in court, but every organisation using or buying AI should document training and reference data sources — licensed, public domain, customer-supplied, or contractual — and test for substantial reproduction of protected works with a fair complaints process.

AI systems learn from vast collections of writing, images, and other material. But being able to find something online does not automatically mean that a company has permission to copy it for every commercial purpose.

Rights holders connected with Elvis Presley's song catalogue filed a copyright lawsuit against Anthropic in California on 23 October 2026. Reporting published over the following days said the action concerns 274 compositions and alleges that protected lyrics were copied without permission to train the Claude chatbot. Anthropic disputes the allegations and intends to defend the case. No court has decided that it infringed copyright in this proceeding.

The legal arguments may address how material was obtained, what rights existed, how the system used the material, and whether an exception to copyright protection applies. Those questions remain open. For ordinary businesses, however, there is an immediate and much simpler issue: can you explain where the material used to build or adapt your AI system came from?

What to do

Every organisation developing, adapting, or purchasing an AI system should be able to identify the sources of important training and reference material. A record should distinguish licensed works, public-domain material, customer-supplied information, and content obtained under contracts. Keep the relevant permissions, limitations, and the name of the person responsible for them.

When working with outside developers or data suppliers, ask for written explanations of where their information came from and what uses are allowed. Check whether contracts address intellectual property claims, notification of disputes, cooperation with investigations, and removal of problematic material. Test whether the AI system reproduces substantial portions of protected works and have a fair process for investigating complaints.

These precautions are not limited to major AI developers. A marketing agency uploading a customer's photographs, an employer using staff manuals to build a chatbot, or a publisher supplying archives to a technology company may each need to consider permissions and confidentiality.

Sources


We link copyright and contract review to data records, supplier checks, and training controls — through gap assessment and AIMS data governance.

Next step

Apply this to your own AI systems

We can tell you in three to four weeks exactly where you stand against ISO/IEC 42001 and the EU AI Act, what carries over from your existing certifications, and what closing the gap would cost.