AI Model Mythos Floods Cyberspace With Unpatched Vulnerabilities; Industry Concedes Patch Limits

Anthropic's Mythos vulnerability-finding AI has exposed a critical inflection point in cybersecurity: researchers can now discover flaws faster than organizations can patch them, potentially making comprehensive remediation impossible. Since launching in April 2026, the model has identified over 134,000 security vulnerabilities across operating systems, web browsers, and open-source projects, with 33,000 or more rated critical or high severity. Among the vulnerabilities Anthropic disclosed to Project Glasswing partners—some 6,157 flaws—fewer than 520 had been patched as of early October, representing roughly 8.4 percent remediation within six months.