If the court approves it, Anthropic will pay at least $1.5 billion to rights holders over how it trained its AI models. This isn’t just a small legal battle — it could impact how much companies pay for AI tools in the future. The lawsuit claims Anthropic used copyrighted works without permission, including pirated books, to train its models.

The case is a big one. The attorneys say it’s the largest copyright settlement in U.S. history. If the court signs off, the money will be split among the owners of the copyrighted works, roughly $3,000 per work. But the judge has raised concerns about the details of the settlement and has delayed approval until certain issues are clarified.

What the lawsuit is about

The lawsuit started with claims that Anthropic downloaded pirated versions of copyrighted books, made copies, and used them to train its AI models. The plaintiffs say Anthropic didn’t get permission to use these works. In court documents, the lawyers called this settlement the biggest of its kind in the U.S.

The court needs to approve the deal, but the judge has questions. He’s worried about missing details like how the works list and claim process will be handled, especially for works with multiple owners. The court has given deadlines for those details to be finalized, with a preliminary approval not expected before late September.

Impacts on AI costs and industry practices

One big concern is how this might affect the price of AI tools. If companies have to pay millions for copyright settlements, they may pass those costs on to customers. Experts say this could lead to a new standard, with licensing fees set at around $3,000 per copyrighted work.

This shift might push AI companies to start licensing their training data upfront instead of risking legal battles later. Barry Scannell, a lawyer, says that new contracts will likely include detailed licenses and warranties about data sources. This could also mean more transparency from data providers like stock photo agencies and music publishers.

Some industry leaders worry that these costs will make it harder or more expensive for businesses to use generative AI. Kevin Hall, a CIO at a credit union, points out that paying for licensed content will be more costly than using pirated data, which is illegal but cheaper. Still, many agree that paying creators fairly is the right thing, even if it raises prices.

What does this mean for AI development and fair use?

Anthropic and many in the AI industry might come out ahead from this case. Jason Andersen, a tech analyst, says the settlement could encourage more legal sourcing of training data. As long as companies pay for content, they might be able to train AI models legally. This could remove a big hurdle for AI development.

However, some experts question whether deleting pirated copies after training makes much difference. Andersen compares it to a hypothetical: if hackers stole data and an AI company used it, should they face a small fine? The precedent set by this settlement might influence future cases about data misuse.

Another issue at play is transparency. Many AI developers don’t tell clients what data they used to train their models. Some universities do try to share details, but most companies prefer secrecy. That leaves enterprises unsure about whether their AI tools are trained on legal or illegal data. Clear information about training data can help ensure models are reliable and compliant with laws.

In the end, this case highlights the ongoing tension between innovation, copyright law, and fair use in AI. It could reshape how AI companies source, license, and disclose their training data, affecting the industry’s future direction.

Inspired by

• https://www.computerworld.com/article/4052812/proposed-1-5-billion-anthropic-copyright-settlement-raises-questions-about-generative-ai-costs.html

Sources

• gmpg.org
• courtlistener.com
• courtlistener.com
• moorinsightsstrategy.com
• cbsnews.com