A US court has approved a $1.5 billion settlement from Anthropic in what is considered the largest copyright infringement case involving AI.
The case began in 2024 when authors Andrea Bartz, Kirk Wallace Johnson, and Charles Graeber filed a proposed class action lawsuit. They accused Anthropic of using their books without permission to train its Claude AI models.
AI models require massive amounts of text for training, with leading models often needing tens of terabytes of data. The authors argued that Anthropic violated their intellectual property rights by using books they owned without their consent.
Judge William Alsup dismissed this part of the case, ruling that training AI models on legally acquired books was “exceedingly transformative” and therefore protected under fair use.
However, Judge Alsup also ruled that using pirated books was illegal. That issue became the focus of a separate trial, which was overseen by Judge Araceli Martinez-Olguin after Judge Alsup retired. She approved the settlement.
A Washington Post investigation revealed that Anthropic ran an internal initiative called Project Panama, which aimed to collect as many books as possible for AI training.
The company reportedly spent tens of millions of dollars buying physical books, scanning them into digital format, and later disposing of the originals. The scanned books were then used to train Claude models.
The legal issue centered on a separate collection of books that Anthropic obtained through piracy. The company used material from Library Genesis (LibGen) and Pirate Library Mirror (PiLiMi), both of which distribute unauthorized copies of books.
Because of this, the court ordered Anthropic to pay about $1.5 billion to affected authors and publishers.
The settlement works out to roughly $3,000 for each illegally acquired book. Lawyers representing the authors had requested 20% of the settlement as legal fees, but Judge Martinez-Olguin reduced that to less than 7%, or about $101 million.
The judge also reduced the awards for the three lead plaintiffs from $50,000 to $15,000 each.
Although this settlement is the largest of its kind, Anthropic continues to face several legal challenges. Reddit recently sued the company, accusing it of using bots to scrape content from its platform even after being told to stop.
Earlier this year, book publisher Chicken Soup for the Soul also sued Anthropic, along with OpenAI, Meta, Apple, and NVIDIA. The publisher claims the companies used its copyrighted books to train AI models without permission.
The ruling also comes weeks after Anthropic accused Alibaba of using Claude to improve its own AI models through knowledge distillation.
Anthropic claimed Alibaba used nearly 25,000 fraudulent accounts and about 28.8 million interactions with Claude to generate outputs that were then used to train its Qwen models.
Separately, a researcher alleged that Anthropic embedded tracking tools in Claude Code to monitor users in China.
Until recently, Anthropic was widely regarded as one of the leaders in AI model performance. However, access to its Fable 5 and Mythos 5 models was temporarily restricted for foreign nationals, including Anthropic employees who were not US citizens.
US export controls cited national security concerns before the restrictions were later relaxed with additional safeguards.
Anthropic is also facing growing competition. OpenAI recently introduced GPT-5.6, Google launched Gemini 3.6 Flash and is expected to release Gemini 3.6 Pro soon, while Moonshot AI released Kimi K3, which has produced benchmark results comparable to Claude in several tests.
