Anthropic unveils Project Glasswing and withholds Claude Mythos Preview

Anthropic announced Project Glasswing on Tuesday, April 7, a new industry effort with AWS, Apple, Google, Microsoft, Nvidia, Palo Alto Networks and others to secure critical software against AI-driven attacks. At the same time, the company revealed Claude Mythos Preview, an unreleased frontier model that Anthropic says has already uncovered thousands of high-severity vulnerabilities across operating systems, browsers and other core infrastructure.

This is not a routine model update. Anthropic says the model is powerful enough on offensive security tasks that it will not be released broadly for now. Instead, a limited partner group will use it for defensive work, while Anthropic commits up to $100 million in usage credits and $4 million in direct support for open source security organizations.

For CIOs, the signal is stark. AI is now shifting the threshold for both attack and defense in cybersecurity, and the window between vulnerability discovery and exploitation is getting shorter. If frontier models can surface flaws that survived decades of human review, security can no longer be treated as a slow, mostly manual side function.

The most interesting part may be the packaging. Anthropic is framing this as a defensive alliance, not a product launch. That says a lot about how seriously top AI labs are taking the real-world impact of the next wave of agentic models.