It looks like the honeymoon phase for Anthropic’s clean, ethical AI image is officially over. Over 100 authors have just hit the tech startup with a fresh, massive $75 million lawsuit, claiming the company shamelessly pirated their copyrighted books to train its Claude AI chatbot.
Illegal use of intellectual property
Authors claim Anthropic scraped thousands of copyrighted works from illegal “shadow libraries” to feed its AI data engine.
According to the new complaint filed in federal court, Anthropic skipped paying for licensing fees or asking for permission. Instead, the AI company relied on illegal data aggregates like “The Pile,” a notorious dataset packed with pirated books, to accelerate Claude’s ability to mimic human writing and generate text.
The shadow library problem
This isn’t Anthropic’s first time getting caught with its hand in the digital cookie jar. The company has historically brandished its image as the “safety-focused” alternative to OpenAI, but the courts are revealing a much messier reality behind how their models actually learn.
The crux of the legal argument comes down to data sourcing. In the world of AI training, text is food. To make a chatbot smart, it needs to ingest billions of words.
Instead of buying those books, the lawsuit alleges Anthropic pulled millions of digitized works from known illegal file-sharing hubs like Library Genesis (LibGen) and Books3.
The genesis of the lawsuit
This whole legal mess actually started back in August 2024 when author Andrea Bartz and a group of writers first dragged Anthropic to court.
Fast forward to a landmark judicial ruling in June 2025, when a judge ruled that training AI on legally purchased material is fair game, but downloading stolen or pirated copies is flat-out illegal.
The court found that Anthropic copied and centrally stored around 7 million pirated books to fuel its training operations and ordered the company to pay the authors a $1.5 billion settlement.
The plaintiffs in this new lawsuit opted out of the 2024 lawsuit and are now seeking way more money per complainant than the previous settlement.
The reality check
For everyday users, Claude isn’t going offline anytime soon. But for Anthropic, the financial bleeding is getting impossible to ignore.
While Anthropic previously tried to argue that AI training falls under “fair use,” the plaintiffs are drawing a very sharp line in the sand. Even if the act of training an AI model is deemed transformative, using an explicitly stolen, pirated copy of a book to do it is a separate, clear-cut case of copyright infringement.
Anthropic remains silent, but the message is loud and clear
Anthropic hasn’t issued a public response to this specific filing yet, but one thing is certain. The wild-west era of AI companies scraping whatever they want from the internet without consequences is officially dead.
After being forced to pay out roughly $3,000 per book to authors in its staggering $1.5 billion class-action settlement, this new $75 million suit proves that independent creators are no longer letting tech giants use their life’s work as free fuel.
Source: Crypto Briefing