There's only so much code: the rise in vulnerability numbers
Frontier AI models like Mythos are systematically mining decades of technical debt in legacy libraries and tools, driving Microsoft’s monthly vulnerability advisories from ~100 to 1,415. Despite a 105% year-over-year surge in CVEs, Known Exploited Vulnerabilities (KEVs) grew only 37%, with just 3 of September’s 975 Microsoft advisories seeing active exploitation, demonstrating that discovery volume does not correlate with weaponization. The acceleration challenges the traditional 90-day coordinated disclosure model, as open-source maintainers and vendors struggle to keep pace, while attackers increasingly rely on AI to parse patches faster than human defenders can apply them. The article argues this spike mirrors historical patterns—when format string vulnerabilities emerged in 2000, they peaked and then declined as the backlog was remediated; AI-driven discovery should follow a similar trajectory once legacy codebases are cleared and AI verification is integrated into build pipelines. Defenders must prioritize exploitability within their environments, deploy behavioral anomaly detection, implement interim mitigations, and embed AI-assisted vulnerability scanning directly into development workflows.