ANTHROPIC BUILT AN AI IT DEEMED TOO DANGEROUS TO RELEASE TO THE PUBLIC Anthropic has spent years positioning itself as the safety-first AI company. Now it has built something it considers too dangerous to put in front of the public. Mythos Preview is a general-purpose AI model that happens to be extraordinarily capable at finding and exploiting vulnerabilities in software. The company says it can identify and exploit zero-day flaws in every major operating system and every major web browser. It found and fully exploited a 17-year-old remote code execution vulnerability in FreeBSD entirely on its own, without any human guidance. The vulnerabilities Mythos identifies have, in some cases, survived decades of human review and millions of automated security tests. Anthropic did not release it to the public. Instead the company launched Project Glasswing, a controlled rollout to approximately 50 industry partners selected specifically to test its defensive capabilities. The idea is to let security professionals understand what AI-powered attacks will look like before actual attackers get the same tools. The national security implications rattled Washington. The Trump administration had designated Anthropic a supply chain risk back in March. Mythos changed the conversation entirely. Pentagon tech chief Emil Michael told reporters the model represents a separate national security moment. Congressional homeland security committees have received classified briefings. The question now is who else is building something like this and not telling anyone. Keywords: Anthropic Mythos AI, AI cybersecurity, zero-day vulnerability AI, Project Glasswing national security