Anthropic's most powerful model has found thousands of zero-day vulnerabilities in every major operating system and browser. Rather than releasing it publicly, the company is deploying it as a defensive weapon through a coalition of tech giants — raising urgent questions about what happens when AI outpaces the humans tasked with securing the world's software.