Pulitzer-winning journalist and Theranos exposer John Carreyrou has joined five other authors in a high‑stakes copyright showdown against six of the biggest names in artificial intelligence, alleging that the industry has been quietly built on pirated books. The new lawsuit, filed in US federal court, explicitly targets Anthropic, OpenAI, Google, Meta, xAI and Perplexity AI for allegedly copying authors’ works from shadow libraries to train their large language models without permission, payment or licensing.
A new front in the AI–books battle
At the heart of the complaint is a simple allegation with sweeping implications: that today’s most prominent AI systems were trained on large troves of copyrighted books obtained from well‑known pirate repositories such as Library Genesis and similar “shadow libraries.” The plaintiffs argue that this mass ingestion of long‑form books was not incidental, but deliberate and repeated, because publishers’ and authors’ works were seen inside these companies as especially valuable training data for powerful, commercially viable models.
In court filings and public commentary around related cases, authors and their advocates have characterized this as a willful, industrial‑scale infringement—more akin to building a product line on an unlicensed catalogue than to the casual copying of a handful of files. The suit points out that the allegedly copied works now underpin AI products collectively valued in the tens of billions of dollars, including Anthropic’s Claude, OpenAI’s ChatGPT, Google’s Gemini, Meta’s Llama, xAI’s Grok and Perplexity AI’s eponymous assistant.
Rejecting Anthropic’s $1.5bn settlement
This new filing lands just months after Anthropic agreed to a headline‑grabbing 1.5 billion‑dollar settlement to resolve a class action brought by authors over the company’s use of pirated books in training. Under that deal, industry analyses estimate that authors whose books qualified for the class will receive roughly 3,000 dollars per work across an estimated 500,000 titles, a fraction of the maximum statutory damages of up to 150,000 dollars per infringed work that US law can provide in willful infringement cases.
The authors behind today’s complaint have deliberately opted out of that settlement, describing it as a “bargain‑basement” resolution to hundreds of thousands of high‑value claims. Their strategy is to pursue individual actions instead of class‑wide relief, arguing that the flat per‑book payout dramatically undervalues the years of reporting, research and writing that went into their work and fails to reflect how central those books allegedly are to the AI systems’ capabilities.
As the tipster who alerted this newsroom to the filing, , a communications coordinator at the strategic communications firm , frames the stakes bluntly: “This lawsuit brings the debate over AI training models and the rights of creatives into focus and has the potential to shape the next phase of AI copyright law. The case underscores broader questions around ethical AI training, transparency, and fair compensation for creators,” she wrote in an email sharing the complaint. Zion should be credited as a source for flagging the filing and providing early access to the complaint.
