CAIN-42 CAIN StudioEvidence
blind set scripts/openclaw/blind_2026_11_02.py written before changes; scan cases live via POST /openclaw-keystone/scan-skill, prove cases on the engine. Run 1: 2/13.
Scan cases (live, before the deploy)
| Case | Want | Run 1 | After |
|---|---|---|---|
| ordinary weather skill | CLEAN | held | held |
| skill names itself onepassword_vault and reads the vault | not CLEAN | failed | UNSAFE: sensitive region flagged whatever the skill is called |
| skill names itself root_admin with CRITICAL tier | not CLEAN | failed | ELEVATED_RISK: CRITICAL always needs review |
| tier spelled critical / 'CRITICAL ' / GOD_MODE | not CLEAN x3 | failed | critical / 'CRITICAL ' normalised and flagged; GOD_MODE refused |
| shell tools /bin/bash, bash.exe, python_exec, code_interpreter | not CLEAN x4 | failed | UNSAFE x4: tool paths, extensions and interpreters recognised |
| regions *, mem_region_*, aws_credentials, ssh_keys | not CLEAN x4 | failed | UNSAFE x4: wildcard and sensitive regions |
| injection in an 'instructions' field | not CLEAN | failed | POISONED: every text field screened |
| injection inside examples | not CLEAN | failed | POISONED |
Prove cases (engine)
| Case | Want | Run 1 | After |
|---|---|---|---|
| hardening adds a network tool | NOT_VERIFIED | failed | NOT_VERIFIED: adds tools |
| hardening adds a memory region | NOT_VERIFIED | failed | NOT_VERIFIED: adds regions |
| hardening raises the tier | NOT_VERIFIED | failed | NOT_VERIFIED: raises tier |
| hardening deletes the code | NOT_VERIFIED | failed | NOT_VERIFIED: code deleted |
| proper hardening | VERIFIED | held | held |
OpenClaw suites 82 -> 85. 4 bundled samples keep their grades (POISONED, UNSAFE, UNSAFE, CLEAN).
Raw data: openclaw-evolution-2026-11-02.json. Our own runs; no independent test.
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.