Beyond automation: Why the surge in AI-driven security vulnerabilities demands human technical advocacy

Spring 2026 marked the dawn of AI-driven security vulnerability reporting, beginning with Anthropic’s Claude Mythos AI model claiming discovery of thousands of high and critical vulnerabilities across open source projects through Project Glasswing on April 7. Within weeks, additional vulnerabilities named Copy Fail, Dirty Frag, and Fragnesia emerged, highlighting significant challenges in responsible disclosure practices at this unprecedented scale. Human technical advocates remain essential to coordinate the response, validate findings, and ensure the open source ecosystem can absorb and address AI-discovered vulnerabilities responsibly.