Anthropic Expands Project Glasswing

Before we begin, I recommend reading: Claude Mythos

Anthropic released a statement today announcing the expansion of Project Glasswing to approximately 150 new organizations across more than 15 countries. Initially, Anthropic launched the program with a cohort of roughly 50 partner organizations. Collectively, participants in this program have used the Claude Mythos Preview model to uncover over 10,000 high- or critical-severity security vulnerabilities since April.

Due to the model’s offensive cybersecurity capabilities, Anthropic has opted against releasing Mythos to the general public at this time. Consequently, each of the 150 new organizations must meet stringent security requirements before they are granted access to the model.

Many of these new partners are essential vendors or nonprofits that maintain foundational codebases heavily relied upon by governments and other global organizations. Anthropic estimates that a successful cyberattack on a typical partner’s systems could catastrophically impact over 100 million people. This staggering risk underscores the urgent need to scan codebases with Mythos to proactively patch vulnerabilities before they are exploited. This initiative aligns directly with Anthropic’s long-term goal of utilizing AI to make all software secure.

Imminent Danger

The tech industry is facing a rapidly approaching threat: highly capable AI models with significant cyber capabilities could soon be widely available. Anthropic anticipates that other AI companies will develop models with capabilities comparable to Mythos Preview within the next 6 to 12 months. If similar models fall into the hands of attackers without the necessary safeguards, the consequences could be devastating and immediate.

General Release

Anthropic is working as quickly as possible to safely release Mythos-level capabilities to the general public. Achieving this requires the implementation of highly robust safeguards—which the company notes do not yet exist within the industry—to prevent the misuse of its offensive capabilities.

In the immediate future, Anthropic plans to continue expanding Project Glasswing. Future expansions will prioritize covering essential infrastructure providers, maintainers of critical open-source software, and safety testers.