A security posture is only as strong as its worst-case behavior under load, and Claude Code's patch record shows a product that repeatedly trusted its own controls past the point those controls remained valid. The deny-rule bypass — confirmed by Adversa AI and SecurityWeek — did not require an attacker to know a secret. It required a command chain long enough to exceed an internal threshold, which routine automated refactoring routinely does. Anthropic patched it, but the patch followed public disclosure rather than preceding it.