Elvis Presley Rights Holders Sue Anthropic Over AI Training and Alleged DRM Circumvention

Owners of Elvis Presley's music have sued Anthropic for copyright infringement, seeking $41 million in compensation. The complaint was filed on October 23, 2026, and Engadget reported the filing on October 26.

The suit alleges that Anthropic trained its AI models on 274 Elvis Presley songs without permission. Training means feeding large sets of examples into a model so it learns statistical patterns from them. Among the works cited are "Don't Be Cruel," "Can't Help Falling in Love" and "A Little Less Conversation."

The plaintiffs are described as stakeholders in Presley's back catalog, meaning his older recorded work. They include Authentic Brands Group and Raleigh Music Publishing, along with multiple trusts and at least one individual.

The complaint goes beyond the training allegation. It accuses Anthropic of bypassing DRM (digital rights management, the encryption and access controls that govern how media can be copied or played) and of removing or altering copyright management information, the embedded metadata that identifies rights holders and usage terms, in connection with 247 songs. Think of DRM as a lock on a recording and metadata as the label attached to it. These are access-control and rights-metadata claims, separate from the question of whether training itself was permitted.

Anthropic said it disagrees with the claims in the lawsuit and intends to defend itself.

The broader context here is that the allegations differ in character from a training-only dispute. They concern how the recordings were obtained and whether protective and identifying layers around them were circumvented or stripped. Courts assessing AI training cases will have to weigh those theories alongside any arguments about the training process itself, and the complaint does not resolve how those questions will be treated.

The $41 million figure is the amount the plaintiffs seek. It is a claim, not an established damages award, and the case will proceed through the usual stages before any figure is tested.

In my view, the more durable question for the industry is provenance. Model developers have generally treated training corpora as a matter of data sourcing, while rights holders have pressed for documented acquisition paths and respect for DRM and metadata. A case built on those mechanics puts engineering practice in front of a court. If the claims progress, the likeliest practical outcome for both sides is tighter documentation of where training material came from and under what terms, along with more licensed catalogs. That would give rights holders a clearer basis for compensation and give model builders a cleaner record showing when their data was lawfully obtained.

For now, the filing sets out specific allegations: 274 songs used for training, 247 works affected by alleged DRM circumvention and metadata alteration, and a demand for $41 million. Anthropic says it disagrees and will defend the case.