On June 2, about a week before Anthropic released its Claude Fable 5 model, it announced it had added 150 new organizations to Project Glasswing, the program that allowed (theoretically) trustworthy companies, governments, and other entities to try out the cybersecurity capabilities of Anthropic’s (theoretically) too-dangerous-to-release AI model, Claude Mythos Preview. This, according to a new report from the Washington Post, was part of the move that pla