Limits

Limit
self-run, no independent review
pattern-based: paraphrased injections get through
name-based tool detection has false alarms (a tool named shell_color_theme counts as a shell tool)
a skill's real behaviour is only partly visible from its manifest and code
AI hardening needs Claude credit; scan and prove are free
deployed 2026-11-02 18:56Z; live re-run of the blind set 13/13

FIRME AI CORPORATION · CAIN-42 is a Byzantine trust fabric. Numbers marked "our own check" were run by us, not by an independent party. Nothing on these pages is a certification.