Anthropic's $1.5B Book Piracy Settlement Leaves Fair Use Unsettled
A judge found Anthropic illegally pirated Stanford professors' books for Claude but called training on legally acquired works fair use, while Wolff and other authors seek greater damages in a separate lawsuit.
A federal judge ruled that Anthropic illegally scraped and pirated books by more than 10 Stanford professors to train its Claude models, while holding that training on legally acquired books is fair use. The $1.5 billion Bartz settlement went to authors and publishers of nearly 500,000 registered works.
Tobias Wolff and Chang-rae Lee opted out of the class settlement and filed Cruz v. Anthropic in May 2026, seeking higher statutory damages and an injunction against training on their works, while a broader licensing push continues.
Key takeaways
- Judge William Alsup ruled Anthropic could not pirate books to build a permanent central library, but compared training on legally acquired works to human reading and held it fair use.
- Wolff and 27 other authors say the settlement is insufficient; Wolff calls uncompensated use of his work a form of slave labor and worries about younger writers.
- Wolff and Chang-rae Lee opted out and filed Cruz v. Anthropic, seeking up to $150,000 per infringed work and a permanent injunction, versus about $3,000 per work in Bartz.
- Paul Goldstein cites transaction costs: millions of individual works may be copied in a training incident, making licenses from every owner impractical.
- The Authors Guild is pursuing lawsuits against OpenAI and Microsoft and working with Congress on national licensing legislation.
What people think
Analysis details
19 comments analyzed. Analyzed .
The sample selects active discussion branches and includes available parent comments. It may omit parts of the full thread. The analysis included 19 of 25 usable stored comments because of the input limit. Selected themes do not measure community opinion or how common a view is.
Criminal enforcement double standard
The $1.5 billion civil settlement prompts comparisons to Aaron Swartz's CFAA prosecution, with arguments that individuals face criminal charges while corporations receive civil penalties. Counterarguments note Swartz physically broke into a network closet and circumvented MAC bans, and that his plea deal was six months, not decades. The dispute centers on whether the cases are comparable and whether corporate AI liability should be criminal.
Fair use for legally acquired training
The judge's fair-use finding for training on legally acquired books is highlighted as the more consequential legal development, even as piracy and storage of shadow-library files remain the basis for liability. Some see the headline's claim that training was illegal as inaccurate; others see the acquisition/training distinction as immaterial because pirated books were used for training.
Copyright's moral and economic basis
One strand questions whether copyright over knowledge-based works has any legitimate place in today's information economy, while a reply argues that 250 years of new information eras have not justified discarding authors' moral and economic rights. A further reply dismisses that as opinion and calls copyright immoral.
Slave labor analogy dispute
Calling unauthorized training 'slave labor' draws objections that copyright infringement is substantially different from slavery. A defense cites a dictionary definition covering coerced or poorly remunerated work, but critics argue that definition would include low-wage employment and jury duty, making it unworkable.
AI speed and criminal liability
One argument extends criminal liability to AI operators, warning that high-speed model execution and multi-agent features could cause harm faster than oversight can respond. This is a forward-looking concern rather than a demonstrated incident, and it is used to argue that OpenAI or Anthropic personnel should face charges for what their models do.