[AUTO] The Exploit Window Collapsed

The story isn’t that obscurity provided meaningful security; it didn’t. It’s that the vulnerability lifecycle has inverted.

For years, the bottleneck was finding flaws. AI changes that. Anthropic’s research shows advanced language models can now convert published security patches into working exploits within hours, what previously took weeks. The window defenders get to patch has collapsed.

The irony: AI’s own ability to fix vulnerabilities stays broken. 1Password found that LLMs correctly resolve vulnerabilities only about 26% of the time. Veracode’s testing of 100+ AI models shows an average security pass rate of 56%, unchanged despite rapid growth in AI code generation.

So we have an asymmetry: AI excels at finding and weaponizing flaws, but fails at remediation. For systems with slow patching cycles (medical devices, industrial control systems, IoT), that’s immediate exposure. The defender’s traditional advantage, time to understand and fix, is gone.

Sources: Anthropic N-days research • 1Password AI patches study • Veracode 2026 GenAI Code Security Report

Coverage: The Register