BlindOracle · Live Audit Teardown

A complete agent audit — graded, anchored on-chain, and verifiable without trusting us

This is not a stripped-down demo. It is the exact audit we sell, run on one of our own production agents so you can inspect every layer before you buy. It complements our fleet-wide self-audit and the 30-agent on-chain proof run — see the methodology for how the score is produced.

1 · Result at a glance

C
🟡 Moderate
Security Posture Grade

Verdict: No critical or high-severity issues. 8 medium findings to remediate. Production-acceptable with fixes — not yet attestation-grade (A).

Risk score  ▼ lower is better4.6 / 10
Trust score  ▲ higher is better54 / 100
0
Critical
0
High
8
Medium
0
Low

How to read the grade

GradeRisk bandMeaning
🟢 A0.0–1.5Attestation-grade — ship + anchor with confidence
🟢 B1.5–3.0Low–moderate — minor hardening
🟡 C ← this audit3.0–5.0Moderate — remediate the medium findings, then re-audit
🟠 D5.0–7.0Elevated — do not expose to untrusted input until fixed
🔴 F7.0–10Critical — do not deploy

An all-A result on a real agent would be a red flag — a credible audit finds things. C with zero critical/high is an honest, healthy starting posture.

2 · The cryptographic receipt

ArtifactValueState
Audit content hashc751ac4ae7b7d4402dee9f86d32fb58f33ebfc5dfe083c87a9e20276e75fb43esigned
ProofOfAuditReport (30105)so-30105-5cc98a0a762dHMAC-verified
ProofOfStateAnchor (30106)so-30106-4cc9ed012e09offline-verifiable
Nostr witnessevent bed99c28… · view on njump ↗✅ confirmed
Base mainnet anchorProofAnchor 0x608287cb…82AFe · tx 0x7defcf8a… ↗ · block 47028733✅ confirmed + read-back verified
Honest note — we never fake a green check: the audit's Merkle commitment is anchored on Base mainnet (our own ProofAnchor, owner 0x2D0B…, anchored root 0bae80ab…) and published to Nostr (relay.damus.io + nos.lol). One gap shown as-is: the legacy Base-Sepolia ProofAnchor is owned by a retired deployer address, so we anchor to the funded mainnet contract instead rather than show a Sepolia check we can't honestly sign.

3 · The 8 findings (count bound into the Merkle root — none hidden, none fabricated)

SevOWASP ASIFindingSample technique IDs
🟡 medASI01Agent Goal HijackingAML.T0051, T1059, T1204
🟡 medASI02Tool Misuse & ExploitationAML.T0053, T1106, T1203
🟡 medASI03Identity & Privilege AbuseT1078, T1548, T1134
🟡 medASI05Unexpected Code ExecutionT1059.004, T1620, T1027
🟡 medASI06Memory & Context PoisoningAML.T0020, AML.T0070, T1565
🟡 medASI10Rogue AgentsT1070, T1562, D3-PA
🟡 medTRAP01Content Injection TrapsAML.T0051, AML.T0067
🟡 medTRAP02Cognitive State TrapsAML.T0020, AML.T0018

Findings map across OWASP ASI Top 10, NIST AI RMF and ISO/IEC 42001 — the same crosswalk we apply to customer audits, walked through in the Agent Audit Evidence Kit.

4 · Verify it yourself — no key, no trust

1. Recompute the audit hash from the report  -> c751ac4a…
2. Verify the proof chain (30105 + 30106)    -> 0 bad signatures
3. Confirm the Base-mainnet anchor on public RPC:
   verifyAnchor(0x0bae80ab…) on 0x608287cb…82AFe  -> true
   or open the tx: https://basescan.org/tx/0x7defcf8a…
4. Confirm the Nostr witness: https://njump.me/bed99c28…
5. 8 findings Merkle-committed with N=8 bound in (completeness + inclusion)

This is the same key-free evidence model behind our verifier and agents-without-surveillance posture.

Get your agent audited to this exact standard — $500 flat

5-business-day turnaround. You receive the signed report, the 30105 + 30106 proofs, a W3C Verifiable Credential, an MCP-I Know-Your-Agent credential, a Base-mainnet on-chain anchor + Nostr witness (as shown above), and a passport attestation you can hand any buyer. Managed-audit pricing is on the pricing page.

Prefer email? [email protected]