Anthropic expands Claude Mythos Preview to ~150 more orgs via Project Glasswing

Anthropic announced the expansion of Project Glasswing, extending access to its restricted Claude Mythos Preview to approximately 150 additional organizations in more than fifteen countries — a roughly 4x widening of an initial ~50-partner cohort focused on critical infrastructure. The official post from @AnthropicAI drew 3,866 likes and 419 reposts.

The headline number is the security yield: Anthropic says Mythos has surfaced more than 10,000 high- or critical-severity software vulnerabilities since launching in early April. That positions the model as both a defensive tool — letting organizations find flaws before attackers do — and, implicitly, a dual-use capability that could accelerate offensive cyber operations if misused.

That tension is why the program remains gated. Anthropic says it will not release Mythos-level cyber capabilities to the general public until it has 'highly robust safeguards' in place. The staged, invite-only rollout is the company's attempt to capture defensive value while limiting proliferation — a notable contrast with the open-weights race happening elsewhere in the industry.

It also dovetails with Anthropic's open-sourcing of a 'defending code reference harness' for AI-powered vulnerability discovery the same week. The skeptical read: 10,000 vulnerabilities is an impressive but unaudited figure, and concentrating powerful exploit-finding capability in a vendor-controlled preview raises governance questions about who decides access — exactly the debate now playing out across the security community.