Anthropic’s Project Glasswing now opens its vault to 150 new partners after early April rollout that exposed over 10,000 high‑severity flaws. The initiative, built around Claude Mythos Preview, lets security teams scan codebases for critical vulnerabilities. Early adopters already leveraged the model to discover and isolate threats before release, and users.

The expansion taps 15 countries, drawing power, water, healthcare, communications, and hardware sectors that were under‑represented before. Most new members are vendors or nonprofits that maintain code relied on by governments worldwide. A single breach in any of these codebases could impact more than 100 million people, amplifying the stakes for national and global security today.

Anthropic released Claude Security, a tool that uses Claude Opus 4.8 to scan and suggest patches, and will share partner‑developed triage tools on request. The company plans to shift focus from flaw discovery to disclosure, fixing, and deployment, aiming to create industry standards for deploying AI‑driven security. Success could lock in a defensive edge for critical infrastructure worldwide.

The program also pushes for safer AI releases. Anthropic acknowledges that within six to twelve months other firms may launch Mythos‑class models without safeguards, raising attack frequency. By expanding partner reach and developing robust security protocols, the company seeks to set a precedent for responsible AI deployment in high‑stakes software ecosystems.