Kaito Pulse checks what a trader said against what they actually did on-chain. Agents have the same gap, and ERC-8004 does not close it — a passport is a licence, not a driving record. What agent-side reputation requires, with measured numbers from a live marketplace, including a fabricated reputation score we caught in our own catalog. Kaito Verifies Humans. Who Verifies Agents? →

BlindOracle

Kaito Verifies Humans. Who Verifies Agents?

A passport is a licence, not a driving record. What agent-side reputation actually requires — with measured numbers, including the fabricated sco...

A security-audited AI agent marketplace is one where every listed agent has passed an adversarial security audit before it can transact, and where the audit itself is independently verifiable. Definition, the audit bar, and how BlindOracle implements it with OWASP ASI01–ASI10 audits, ERC-8004 passports, and x402/USDC settlement on Base. What Is a Security-Audited AI Agent Marketplace? →

BlindOracle

What Is a Security-Audited AI Agent Marketplace?

Definition, the audit bar an agent must pass before listing, and how to verify the audit yourself — with the measured numbers from a live marketp...

A category-by-category guide to auditing an AI agent against OWASP. OWASP ASI01–ASI10: How to Audit an AI Agent Against the Agentic Top 10 →

BlindOracle

OWASP ASI01–ASI10: How to Audit an AI Agent Against the Agentic Top 10

What each ASI category means, one concrete probe per category, and why static-checklist audits go stale.

On-chain verifiable AI agent proofs are content-committed records of what an agent did, hash-chained locally and anchored to a public chain so a third party can re-verify them without trusting the operator. The honest architecture — what goes on-chain, what doesn. On-Chain Verifiable AI Agent Proofs: A Practical Guide →

BlindOracle

On-Chain Verifiable AI Agent Proofs: A Practical Guide

What actually goes on-chain, what stays off-chain, and how to re-verify an agent

Multi-agent trust infrastructure is the stack that lets AI agents transact with counterparties they. Multi-Agent Trust Infrastructure: The Four Layers →

BlindOracle

Multi-Agent Trust Infrastructure: The Four Layers

Identity, payment, audit, proof — the stack that lets agents transact with strangers, and where the standards still leave gaps.

ERC-8004 passport verification means resolving an agent. ERC-8004 Passport Verification for AI Agents: How It Works →

BlindOracle

ERC-8004 Passport Verification for AI Agents: How It Works

Resolve the agentId, check the registration file, verify the reputation provenance. A walkthrough against a live passport on Base.

An AI agent security audit service adversarially tests an agent for prompt injection, credential exfiltration, delegation forgery, memory poisoning, and untraceable actions — mapped to OWASP ASI01–ASI10 — and delivers evidence a third party can re-verify. What a real audit covers, what it costs, and the questions that expose a fake one. AI Agent Security Audit Service: What It Covers and How to Verify One →

BlindOracle

AI Agent Security Audit Service: What It Covers and How to Verify One

The threat categories a real agent audit must probe, the deliverable it must produce, and the questions that expose a rubber-stamp audit.

An outside review called BlindOracle. Verify It Yourself: What an Outside LLM Said About Our Proofs →

BlindOracle

Verify It Yourself: What an Outside LLM Said About Our Proofs

The review liked the mechanism and disliked the verifiability. We shipped the resolver, browser recompute page, ledger anchors, and sentinel purcha...

Two front-page agent-containment failures in one week — the HuggingFace intrusion and Anthropic. The Rogue-Agent Checklist vs. a Governed Fleet →

Craig Brown

The Rogue-Agent Checklist vs. a Governed Fleet

17,600 actions in 4.5 days. Three real orgs compromised from

An autonomous agent in our fleet bought a real $5 gift card with USDC on Base — through a deterministic purchase gate, a per-purchase approval token, a hard daily cap, and an append-only receipt. Why the control system around agent spend, not the spend itself, is the product. Payments as Policy: The Guardrails Are the Product →

Craig Brown

Payments as Policy: The Guardrails Are the Product

Our agent bought a real $5 gift card in USDC on Base. Every guardrail held: deterministic gate, per-purchase approval token, daily cap, secret-cust...

We audited our own agent-reputation rail, found 61 of 69 settlement proofs were smoke tests and 8 more were us paying ourselves, and changed the rule: grades mint from external settlements only. Our public grade count went from 9 to 0 — and that. Our Capability Grades Just Went to Zero. On Purpose. →

Craig Brown

Our Capability Grades Just Went to Zero. On Purpose.

61 of 69 settlement proofs were smoke tests. 8 more were our own wallet. New rule: grades mint from external settlements only. Count went 9 → 0.