AI Vulnerability Discovery Is Outpacing Remediation

This report examines a rapid inflection in AI-driven vulnerability discovery and exploitation: modern models (e.g., Opus 4.6, Mythos) are autonomously finding and weaponizing many more flaws across major open-source projects than traditional methods did, revealing that vulnerabilities may be far denser than previously believed. It cites competition and testing data (large jumps in identification/patching rates, hundreds of findings and dozens of working exploits against Firefox and OSS-Fuzz targets) and warns that remediation processes designed for a sparse discovery era will be overwhelmed; it recommends organizational redesign, feedback to secure design, and high-fidelity validation to prioritize exploitable risk.