Anthropic has expanded Project Glasswing, the controlled-access initiative that gives selected organizations use of its unreleased Claude Mythos Preview model for cybersecurity testing. The company is adding approximately 150 new partners focused on critical infrastructure and widely used open-source code.

Initial partners had already reported finding more than 10,000 high- or critical-severity vulnerabilities. The expansion covers organizations in more than 15 countries and additional sectors including power, water, healthcare, and communications.

Each new participant must meet Anthropic’s security requirements before receiving access. The program remains separate from general public model releases. Related work on AI infrastructure and content systems is covered in the creator stack overview.

Key takeaways

What the Glasswing expansion changes for security teams

Mythos Preview is positioned as a frontier model with strong vulnerability-discovery capabilities. Access stays restricted to vetted defenders rather than broad release. Anthropic has also merged related programs into a tiered Cyber Verification Program structure with Defense, Red Team, and Specialized Access levels.

Open-source maintainers have noted the volume of findings can outpace patching capacity. Some have requested slower disclosure cadence while remediation catches up. The company continues to allocate model credits and direct support for the effort.

What to watch next

Further geographic and sector expansions are planned. Public reporting on total validated findings and average time-to-patch from the larger cohort will indicate whether the program scales beyond the initial 10,000-flaw milestone.