Anthropic Expands Project Glasswing to 150 New Critical Infrastructure Partners

Anthropic is scaling Project Glasswing, its program that grants vetted organizations access to the Claude Mythos Preview model for vulnerability scanning, from roughly 50 initial partners to about 150 new ones across more than 15 countries. The expansion deliberately pulls in sectors the first cohort missed — power, water, healthcare, communications, hardware — along with upstream software vendors whose codebases underpin governments and other infrastructure operators. Partners so far have surfaced more than 10,000 high- or critical-severity flaws, and Anthropic estimates a successful attack on a typical participant could affect over 100 million people.

The pitch is that cheap, capable cyber-offensive models from other labs are 6 to 12 months out, possibly without safeguards, so defenders need a head start. Beyond finding bugs, partners are now using Mythos Preview to write patches, run pre-release checks, automate pen testing, and port legacy code to memory-safe languages. Anthropic is also shipping Claude Security — a scan-and-patch product built on Opus 4.8 — and selectively releasing internal Glasswing tooling to trusted security teams.

The company concedes the real bottleneck has shifted from discovery to verification, disclosure, and patching at scale, particularly for open-source maintainers, and says it is in talks with third parties about industrializing that pipeline. General release of Mythos-class capabilities is gated on misuse safeguards that Anthropic admits neither it nor any peer has yet built, so the near-term plan is further controlled expansion via Glasswing and a forthcoming Cyber Verification Program.