A federal judge in San Francisco has ruled that Anthropic’s training of Claude on legally purchased and digitized books qualifies as fair use — a landmark victory for AI models and a potential legal shield for the industry . The court recognized that ingesting and distilling these texts to train transformative systems aligns with copyright’s intent “to foster scientific progress” .

But the celebratory headlines mask an important caveat: Anthropic still faces a December trial over the alleged piracy of over 7 million books sourced from shadow libraries — a move deemed unlawful and not protected as fair use .

This ruling sends a clear signal: AI training through legitimate means is OK, even encouraged — but innovation built on stolen content won’t stand. A milestone for generative AI, but with a hard reminder: how you build your model matters as much as what it does. 👨‍⚖️📚