Zscaler Joins Project Glasswing to Harden Infrastructure

Zscaler is publicly joining Project Glasswing, Anthropic's restricted initiative, and received access to Claude Mythos Preview to test AI-driven vulnerability discovery against real-world attack surfaces. Zscaler highlights that Mythos Preview can chain logic flaws and generate working exploits at machine speed, surfacing thousands of high-severity flaws during preliminary evaluations.

Claude Mythos Preview is described as a frontier code-understanding model that maps program logic, identifies multi-step exploit chains, and synthesizes proof-of-concept exploits far faster than manual analysis. Zscaler emphasizes three practical consequences:

• •It scales parallel scanning of internet-facing assets beyond human capacity, enumerating complex attack graphs.
• •It automates exploit synthesis by linking disparate weaknesses into working chains rather than flagging isolated issues.
• •It reduces time-to-exploit from weeks to hours, compressing the defensive window.

The collaboration signals a turning point for defenders: powerful LLM-style code models are now offensive-grade tools for vulnerability discovery. Zscaler frames this as a systems and architecture problem rather than a tooling-only problem. The post criticizes the thirty-year perimeter model, arguing that firewalls, VPNs, and patching workflows assume human-speed attackers and therefore cannot scale against automated model-driven reconnaissance.

Security teams must prioritize exposure minimization and least-exposure architectures for applications and APIs. Traditional investments in incremental scanning or signature detection will be insufficient as automated adversaries dramatically shorten the window between discovery and exploitation. Expect increased emphasis on design patterns that remove public-facing surfaces, stronger identity- and policy-based access, and proactive threat modeling that accounts for AI-accelerated attack techniques.

Watch for practical tooling and frameworks that codify least-exposure patterns, and for more defenders to request controlled access to models like Claude Mythos Preview to harden critical systems. Also monitor disclosure and coordination practices as AI finds larger volumes of high-severity issues.