Judge Alsup ruled that model training was lawful but penalized Anthropic’s use of illegal shadow libraries.
Judge William Alsup ruled that Anthropic’s training on copyrighted books is lawful, but ordered the company to pay $1.5 billion for obtaining books from illegal shadow libraries. He compared large language model training to a writer studying literature, not to copying a work.
The ruling comes from Judge William Alsup in a case brought by authors against Anthropic. Attorney Cathy Gellis argues the decision is generally good news for AI companies, because copyright law hinges on copying, not on reading or consuming a work. The law has not been updated since 1976, leaving judges to interpret outdated guidelines for new AI questions.
For AI teams, the ruling separates lawful training from lawful sourcing. Building models on openly licensed or public domain data avoids the risk of copyright liability. Enterprises should audit training data provenance and stay informed on fair use rulings, as the legal landscape remains unsettled.
Copyright law has not kept pace with AI, and appellate courts will shape the boundaries of fair use. Future cases will decide whether training on large-scale web data is transformative. Operators should monitor decisions from higher courts and adjust data acquisition strategies accordingly.
What matters
- Judge Alsup ordered Anthropic to pay $1.5B for pirated books, but ruled training lawful.
- Builders must verify training data provenance to avoid copyright liability themselves.
- Watch for appellate court decisions that will define fair use for AI training.
Why it matters
Watch for appellate court decisions that will define fair use for AI training.
This GenAI News article was prepared in original wording using reporting and materials published by TechCrunch AI. Source reference: https://techcrunch.com/2026/08/23/is-it-legal-to-train-ai-models-on-copyrighted-books-its-complicated/.
Drafted by the GenAI News review pipeline.