Anthropic built something it won't release. The company's new Claude Mythos Preview model is so good at finding software vulnerabilities that it's being kept from the public entirely. We're talking tens of thousands of zero-day exploits across every major OS and browser, things top human security researchers would miss. Instead of a public launch, Anthropic is giving access to about 40 organizations through a new initiative called Project Glasswing, backed by $100M in credits.
Microsoft, Apple, CrowdStrike, and Amazon are in. Everyone else waits. It's a weird moment for AI, when the company that built the thing says "actually, no." And it raises a question nobody really wants to answer: what happens when the next lab doesn't hold back?