Anthropic faces lawsuit seeking to seize and destroy AI model weights

The Elvis Presley copyright claim

Anthropic is currently facing a lawsuit involving 274 compositions associated with Elvis Presley. The plaintiffs claim the company copied songs without authorization to teach its AI model, Claude, while also deleting copyright-management information like publisher data.

The lawsuit seeks $48.6 million in total damages. This figure includes $41.1 million in statutory copyright damages and an additional $7.535 million under the Digital Millennium Copyright Act, as reported by Music Business Worldwide.

An Anthropic spokesperson stated, "We disagree with the claims and intend to defend this case fully." The company maintains that it has the legal rights or protections to use its training data, including the argument that such uses may qualify as fair use.

Reaching inside the model weights

What makes this case unique is the specific remedy the plaintiffs are requesting from a federal court. They are asking for the seizure or destruction of not just training datasets, but also internal repositories, server infrastructure, and model weights.

Model weights are the numerical values that an artificial intelligence system learns during its training process. These values are essential because they help determine how Claude reacts to different inputs and prompts provided by users.

While this does not mean a court will automatically destroy Claude, the request highlights a specific type of legal risk. It suggests that copyright remedies could potentially penetrate the very architecture of the AI rather than just resulting in a monetary payout.

Risks for prospective investors

Anthropic recently filed a registration statement with the Securities and Exchange Commission to sell shares to the public. In this filing, the company explicitly cautioned that unfavorable copyright rulings could limit its ability to distribute impacted models.

The prospectus notes that a negative court decision could force Anthropic to retrain its models at a significant expense. This is a major concern because rebuilding a frontier AI model requires massive amounts of processing power, engineering labor, and time.

Investors are looking closely at these risks because Anthropic's growth is tied to its ability to compete with rivals like OpenAI and Google. The company has already shown the scale of these costs by settling a previous copyright lawsuit involving writers for $1.5 billion.

The broader legal landscape

The Presley case is just one piece of a larger legal puzzle regarding how contemporary copyright laws apply to AI training. Courts are still figuring out whether the content used to create today's most useful models should have been used in the first place.

One flaw in the current complaint is that the plaintiffs do not specify a single instance of Claude reproducing a Presley song. They allege that the model created infringing material, but accusations are not the same as legal conclusions.

Anthropic may be able to refute some or all of these allegations as the case moves forward. However, the litigation serves as a timely illustration of the risks the company identified in its own public offering documents.

What comes next for AI law

The outcome of this case could set a precedent for how courts handle the 'materiality' of copyrighted data within a trained model. If a judge determines that certain data shouldn't have been used, it could lead to injunctions that restrict access to current models.

This could lead to a situation where AI companies must constantly monitor and potentially prune their models to avoid legal liability. Such a requirement would increase the complexity and cost of maintaining high-performing AI systems.

For now, the focus remains on whether the court will allow copyright remedies to reach inside the model weights. The result will determine if AI developers can continue using current training methods or if they will be forced to find new ways to build their technology.