← Back to feed

Project Glasswing is expanding. Anthropic announced on June 2 that it is adding approximately 150 new organizations to the vulnerability scanning program, bringing total participating partners to roughly 200. The new cohort spans more than 15 countries; Anthropic says future expansions will push the geographic footprint further.

The original April cohort was technology-heavy. The new wave fills the gaps: power utilities, water infrastructure, healthcare systems, telecommunications providers, and hardware manufacturers. Many of the new partners are software vendors whose codebases sit upstream of thousands of other organizations — governments and enterprises included. Anthropic’s selection criterion is blunt: admission requires that a major attack on the partner’s codebase would plausibly affect more than 100 million people.

The patching bottleneck

Glasswing’s first six weeks produced more than 10,000 high- or critical-severity vulnerabilities. The current challenge is not finding flaws — Claude Mythos Preview does that at scale. The bottleneck is everything that comes after: verifying findings, coordinating disclosure, writing and testing patches, deploying fixes.

Anthropic is addressing this directly. Glasswing partners can now use Mythos Preview not just to scan codebases but to draft patches and run pre-release checks. The program also provides shared tooling — released on request to trusted security teams — that participating organizations developed during the first phase to speed triage.

The company also released Claude Security, a public-facing product that uses Claude Opus 4.8 to scan codebases and suggest fixes. It sits below the Mythos Preview access tier, available to organizations that don’t qualify for full Glasswing membership.

The 6-to-12-month window

Anthropic frames Glasswing as a race against a near-term shift in the threat landscape. The argument: within six to twelve months, multiple AI companies will have models comparable in capability to Mythos Preview, and some will release them without the access controls and oversight Anthropic has built around its own deployment.

That scenario — Mythos-class capability available without safeguards — is the policy backdrop for the expansion. Anthropic is trying to move critical infrastructure defenders up the learning curve before offensive capabilities diffuse through the market. The 150 new organizations are selected because they have the most to lose and the widest blast radius if they are not ready.

The order of operations Anthropic is working toward: first, make scanning universal; second, accelerate patching to match the pace of discovery; third, reshape industry norms around what pre-deployment security looks like when AI can both find and exploit vulnerabilities in automated loops.

Key facts

• Total Glasswing partners: ~200 (50 original + 150 new)
• Countries covered: 15+, with further geographic expansion planned
• New sectors: power, water, healthcare, communications, hardware
• Selection threshold: major attack would affect 100M+ people
• Mythos Preview remains the scanning engine; Claude Opus 4.8 powers the public Claude Security product
• Anthropic’s forecast: Mythos-class models from other labs within 6-12 months