Authors sue Anthropic over pirated books

Even the "safety-first" AI darling isn't immune to the copyright reckoning.

A group of authors just filed a proposed class action lawsuit against Anthropic in a California court. The filing doesn't mince words, accusing the startup of building a multibillion-dollar business on the backs of hundreds of thousands of stolen, copyrighted books.

The ghost of Books3 strikes again

The lawsuit zeroes in on a notorious piece of training data known as Books3, which was packed inside the massive open-source dataset called "The Pile."

The authors leading the charge include Andrea Bartz (We Were Never Here), Charles Graeber (The Good Nurse), and Kirk Wallace Johnson (The Feather Thief). They claim Anthropic used The Pile to train its flagship Claude family of models.

Books3 is essentially a pirated digital library, packed with ebooks scraped from torrent sites like Bibliotik, featuring high-profile authors like Stephen King and Michael Pollan. According to the complaint, Anthropic knowingly downloaded and reproduced this copyrighted material.

Making matters worse, Anthropic basically handed the plaintiffs ammunition earlier this month when it confirmed to Vox that it did, in fact, use The Pile to train Claude.

The authors are asking the court for class certification, statutory damages, and an injunction blocking Anthropic from using copyrighted works in future training runs.

A familiar legal battlefield

While Books3 was eventually stripped from the official release of The Pile, the lawsuit notes that the original dataset is still floating around online.

Anthropic isn't the only one taking heat for The Pile. Recent reports revealed that Anthropic, Apple, and other tech giants also relied on scraped YouTube video subtitles tucked inside the same dataset.

The lawsuit mirrors legal battles already raging across Silicon Valley. Meta, Microsoft, and EleutherAI were hit with a similar suit by former Arkansas Governor Mike Huckabee, while George R.R. Martin and Jodi Picoult have OpenAI tied up in court over identical allegations.

Anthropic did not immediately comment on the filing.

Why it matters

This is a massive headache for Anthropic, arguably more so than the suits OpenAI is fighting.

For starters, Anthropic's brand is built around being the cautious, safety-conscious alternative to OpenAI. Getting dragged into court for hoovering up pirated ebooks directly contradicts that white-hat narrative.

More importantly, we think Anthropic put itself in an extraordinarily weak legal position by publicly confirming to Vox that it trained on The Pile. Most AI companies try to hide behind proprietary obscurity during the discovery phase, but Anthropic effectively acknowledged having the exact dataset in question.

If courts rule that using pirate-sourced open-source sets like Books3 constitutes willful infringement, the financial damages could be staggering. Beyond money, the nuclear outcome would be model disgorgement—forcing Anthropic to delete weights trained on that data.

At the very least, this lawsuit will force Anthropic to follow OpenAI's lead and spend tens of millions of dollars licensing legitimate media and book catalogs, permanently raising the barrier to entry for training frontier models.