A new mid-year legal review reveals a major shift in the AI copyright wars of 2025. While courts in California recently ruled that training AI models is largely fair use (Bartz v. Anthropic, Kadrey v. Meta), the industry is facing a new, existential threat: Digital Piracy. Judges have ruled that using pirated datasets (like shadow libraries or torrents) is likely not fair use, potentially exposing companies like Anthropic and Meta to billions in statutory damages. The report also details the massive new lawsuits filed this year, including Disney v. Midjourney and a class action by adult content producers against Meta.