The Story

Anthropic has announced Project Glasswing, a cybersecurity initiative built around Claude Mythos Preview, an unreleased frontier AI system the company says is powerful enough to warrant restricted deployment.

The initiative launches with partners including AWS, Apple, Google, Microsoft, Nvidia, and several cybersecurity organizations, positioning Mythos as a specialized model for defensive security research rather than public use.

What Anthropic claims Mythos can do

• Advanced vulnerability discovery: Anthropic says the model identified thousands of security flaws across major operating systems and browsers.
• Legacy bug detection: Some vulnerabilities reportedly survived decades of prior review and automated scanning.
• Broad capability jump: Internal benchmarks show performance improvements over existing Claude and competing frontier models.

Restricted rollout

• No public release: Anthropic is not making Mythos broadly available at this stage.
• Limited access: Only selected launch partners and vetted organizations will receive access.
• $100M support pool: Anthropic is backing defensive security work with substantial credits/resources for partners.

Why Anthropic says it matters

Anthropic frames Glasswing as preparation for a future where frontier AI models may become powerful enough to discover and exploit vulnerabilities faster than current security systems can handle.

The company says defensive infrastructure and coordination must mature before such systems become widely accessible.

Why it matters

This is a notable sign that frontier labs are beginning to withhold certain model capabilities from public release when they believe the capability/risk balance is too sensitive.

It also offers a glimpse into the type of highly capable internal systems major AI labs may already be developing beyond what is publicly available today.