Anthropic expands Project Glasswing after early partners find 10,000 severe flaws

Anthropic's Project Glasswing expansion turns model capability into a concrete security program. The company says the initial group of roughly 50 partners has been using Claude Mythos Preview to scan codebases and has found more than 10,000 high- or critical-severity flaws. Anthropic is now extending access to approximately 150 new organizations that must meet security requirements before using the model. The new group spans more than 15 countries and mostly includes organizations that support critical infrastructure. The announcement matters because it is not a generic model launch: it shows frontier cyber models being distributed through controlled partnerships, with measurable vulnerability-discovery output, while governments are simultaneously asking how to vet advanced AI cyber capabilities.

Key details: June 2, 2026, Project Glasswing, Claude Mythos Preview, roughly 50 initial partners, more than 10,000 high- or critical-severity flaws, approximately 150 new organizations, more than 15 countries, critical infrastructure.

Continue swiping for more AI Brief stories.