Independent security audit — open RFP.
The largest gap in the claim ledger is that no third party has audited CAIN-42. This RFP defines a bounded, lawful, reproducible engagement: try to produce an unauthorized effect, or make an authorized action look different in the evidence. Publish what you find.
⬇ Full RFP (Markdown)Scope JSONSubmit a proposal
The one question
Can an unauthorized action produce a real effect, or can an authorized action be misrepresented in the evidence? Everything else is secondary.
In scope
- Boundary bypass: a real tool effect without a valid, unexpired, unrevoked, single-use ALLOW bound to that exact tool, arguments and tenant.
- Grant forgery, replay, expiry, revocation, cross-process reuse, stale fencing and budget exhaustion.
- Child/delegation authority escalation.
- Cross-tenant memory, task, effect and evidence isolation.
- Making memory, model output, reputation or a skill become authority.
- Evidence modify / remove / add / reorder / truncate / re-sign; receipt vs observed effect disagreement.
- Duplicate effects after lost acknowledgement, worker kill, lease expiry, host reboot or failover.
- Secret leakage and supply-chain provenance.
- Bounded denial of service on the isolated replica only.
Out of scope
- Model quality, benchmarks and “AGI” claims.
- Cloud-provider internals, host kernel and physical security.
- Hardware attestation / TEE — none is deployed; do not assume one.
- SOC 2, ISO 42001 or EU AI Act certification.
- The quarantined PBFT cluster for customer anchoring.
Rules of engagement
- No production access. Do not test cainstudio.online, mcpgate.online, clawx.click, the live cluster, customers or third parties.
- No customer data. Synthetic tenants and the development replica only.
- No destructive tests outside the replica; no shared-infrastructure DoS.
- No secret exfiltration. Record location and a redacted proof.
- Private report first to firmeai2026@gmail.com; 90-day disclosure by default.
- Safe harbour for good-faith research that follows these rules.
Indicative budget (not an offer)
| Phase | Scope | Range | Time |
|---|---|---|---|
| Focused boundary review | 8 attack classes, 5 auditor-days | $15k–$30k | 2 weeks |
| Full audit + recovery + evidence | 9 classes, replica drills, retest | $45k–$90k | 5–7 weeks |
| Retest only | Verify fixes | $8k–$15k | 1 week |
Deliverables
- Executive summary, findings with severity and exact reproduction, fix verification, coverage statement.
- A public summary we can publish with the claim ledger, and a signed independence attestation.
- Unresolved findings stay open in public.
Proposal
Email firmeai2026@gmail.com with subject “E82 audit proposal”. Include firm, lead auditor, relevant agent/MCP/distributed-systems work, methodology, independence statement, work plan, price and a redacted sample report.
Issuer: FIRME AI CORPORATION · source commit 9bb134a72b0f7af68d02071adaf30b0cf1e39bab · bundle Merkle 63868034… · contact firmeai2026@gmail.com.