Anthropic’s Mythos Finds 10,000 Security Flaws, Exposes Patching Obstacles

Anthropic’s Mythos AI model and Project Glasswing uncovered tens of thousands of vulnerabilities (thousands rated high/critical) across vendors and open-source projects, demonstrated the ability to generate exploits (e.g., CVE-2026-5194 in wolfSSL), and prompted major vendors to increase patching activity; the report warns this capability will force changes to disclosure, verification, and patching processes to mitigate heightened exploitation risk.