Anthropic’s Project Glasswing has identified over 10,000 high- and critical-severity software vulnerabilities utilizing its Claude Mythos Preview AI model. This initiative aims to leverage advanced AI coding technologies for defensive cybersecurity measures, established to mitigate risks before the potential misuse of the model occurs. Project Glasswing operates as a controlled-access program in collaboration with significant technology and security organizations to enhance and secure vital software infrastructure.

Anthropic: Anthropic is an AI safety and research company focused on building reliable and steerable AI systems. It recently launched Project Glasswing as a defensive cybersecurity initiative that applies its advanced frontier models to identify vulnerabilities in critical software. The company has partnered with major technology providers to make the model available selectively for hardening essential codebases.
Project Glasswing: Project Glasswing is an Anthropic-led coalition launched in April 2026 to secure the world’s most critical software through AI-assisted vulnerability discovery and patching. It provides controlled access to a powerful unreleased model for defensive work by vetted organizations and open-source maintainers. The initiative emphasizes proactive security measures in response to emerging AI capabilities in cybersecurity.
Claude Mythos Preview AI model: Claude Mythos Preview is an unreleased general-purpose frontier model developed by Anthropic with advanced capabilities in software engineering, reasoning, and vulnerability analysis. It is not being made publicly available due to its potential for both beneficial and harmful cybersecurity applications. The model is deployed exclusively through Project Glasswing for defensive scanning of critical software.

Initiative Purpose: Project Glasswing was created to harness advanced AI coding abilities for defensive cybersecurity work ahead of any potential public release or misuse of the underlying model.
Partnership Approach: The project operates as a controlled-access program involving major technology and security organizations to scan and harden essential software infrastructure.