Limits
| Limit |
|---|
| self-run, no independent review |
| pattern-based injection screen: paraphrases get through |
| tool mentions in messages are matched by name only |
| it audits the declared topology, not what agents actually do at run time |
| AI hardening needs Claude API credit; the deterministic scan and prove are free |
| deployed 2026-11-02 18:56Z (gateway rolling deploy); live re-run of the blind set: 12/14 = the 1 by-design case + the 1 run-1 case invalidated by test design (its corrected run-1b version holds) |
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.