“The National Security Agency has been testing Anthropic's Mythos AI model to find cybersecurity vulnerabilities in software, including Microsoft products.”