In a landmark ruling, a federal judge determined that Anthropic illegally scraped copyrighted books from "shadow libraries" such as LibGen and PiLiMi to train its Claude large language models (LLMs), violating copyright laws in a May 2026 lawsuit (Bartz v. Anthropic). Judge William Alsup blocked the practice of pirating books for a permanent central library but sanctioned the legal acquisition of texts for training, citing fair use principles. He analogized AI training to human reading, arguing that LLMs "internalize" works to generate novel outputs rather than replicate them. The $1.5 billion settlement compensated authors and publishers of nearly 500,000 registered works, though some plaintiffs, including Stanford professors Tobias Wolff and Chang-Rae Lee, rejected the deal. They filed a separate lawsuit (Cruz v. Anthropic) seeking up to $150,000 per infringed work, alleging unauthorized downloading and storage of pirated files. Legal scholars highlighted "transaction costs" as a barrier to licensing millions of works individually. Over 70 lawsuits against AI firms, including cases by The New York Times and Getty Images, underscore ongoing disputes over mass data scraping. The Authors Guild advocates for mandatory licensing frameworks to ensure compensation and attribution, while authors like Wolff emphasize protecting future writers’ rights. The rulings reflect tensions between copyright law and AI development, with no clear resolution yet on mass scraping practices.