Anthropic is broadening its Project Glasswing cybersecurity initiative, adding approximately 150 new partner organizations as it prepares wider releases of its Claude Mythos AI model. The controlled rollout that began in April with select partners like Apple will now extend to around 200 total organizations, each required to meet strict security requirements before gaining access.

The expansion includes partners across more than 15 countries, with Samsung and NATO reportedly joining the effort. Anthropic notes that communications and hardware companies are better represented in this latest wave, reflecting the cybersecurity-focused nature of the program. These partners will work alongside open-source software maintainers and US government entities in the collaborative security initiative.

Meanwhile, Claude Mythos-class models are slated for general availability within weeks. Anthropic emphasized that models of this capability level require enhanced cyber safeguards before public release, and the company is making rapid progress on implementing these protections. The timing aligns with Anthropic's announced plans to pursue public company status later this year.

This expansion suggests Anthropic is prioritizing security validation before unleashing its most powerful AI models broadly, potentially setting a precedent for how advanced AI systems reach market while maintaining safety standards.