Anthropic Trains Claude with Stolen Stanford Professors' Books
anthropic claude
| Source: HN | Original article
Anthropic used Stanford professors' books without permission to train its Claude AI model.
A federal judge has ruled that Anthropic illegally scraped and pirated the books of more than ten Stanford professors to train its Claude language models. The decision, issued in the May 2026 case Bartz v. Anthropic, confirms that while the company may use works it lawfully acquires under a fair‑use standard, it cannot build a permanent “central library” from pirated material. The ruling clears the way for a $1.5 billion settlement that covers roughly 500,000 registered works, but a subset of authors – including Tobias Wolff and Chang‑rae Lee – have launched a separate suit, Cruz v. Anthropic, seeking up to $150,000 per infringed title.
The judgment matters because it sharpens the legal line between permissible data mining and copyright infringement for AI developers. Anthropic’s reliance on a massive, unlicensed corpus underscores a broader industry practice of harvesting text from shadow libraries, a method that has been defended as essential for scaling large language models. By deeming the piracy illegal, the court signals that future AI training pipelines will face tighter scrutiny and may need to secure explicit licences or rely on clearly fair‑use‑eligible sources.
Watch for Anthropic’s response to the settlement and the outcome of the Cruz case, which could set a higher per‑work damages benchmark. Regulators in the U.S. and Europe are already probing AI data‑use practices, so further litigation or legislative action may follow. The decision also puts pressure on other AI firms to audit their training datasets, potentially reshaping how the sector sources copyrighted material and influencing the cost and speed of model development.
Sources
Back to AIPULSEN