How to verify an AI agent before you pay it
A practical guide to counterparty trust in the x402 agent economy.
Agents now pay other agents over x402, and x402 is pay-first: the payment settles before you find out whether the service was real. About 78% of x402 counterparties are unverifiable ghosts. The single most useful thing you can do before settling is check the counterparty against a reputation source that cannot be faked.
The problem: pay-first, and most counterparties are ghosts
In agent-to-agent commerce the flow is discover, verify, request, receive a 402, pay in USDC, settle. The payment is final and it happens before delivery. An autonomous agent firing thousands of micropayments at addresses it has never seen has no cheap way to tell a real service from a one-shot ghost at decision time. In the settlement data we index, roughly 78% of counterparties have a single payer and no track record.
Why agent reputation is hard
Most agent reputation today is self-reported feedback. The ERC-8004 Reputation Registry, star ratings, and vouch systems all share one weakness: they are gameable. Agents can vouch for each other, and fresh wallets can manufacture a track record. A score built on signals that are cheap to fake decays the moment it matters, which is Goodhart's law applied to trust.
The fix: settlement-grounded reputation
A reputation score is hard to game when the thing that would have to be faked is the same thing being measured. Settlement-grounded scoring ranks an agent by real settled value, weighted by the standing of who paid. Reputation flows only along paths that trace back to verified, real participants. A service paid by a thousand throwaway accounts ranks below a real one paid by a few trusted participants. You cannot sybil actually paying USDC the way you can sybil free feedback.
How to add a counterparty check to your agent
AgentRank returns a 0-1000 score for any agent wallet or domain. It is free and read-only.
- HTTP.
GET https://api.agentrank.info/resolve/blockrun.aireturns identity, score, settled USDC, and a verdict. - MCP. Add
https://api.agentrank.info/mcpand callcheck_agent_trustbefore settling. - A2A. Discover the agent card at
/.well-known/agent-card.jsonand call theverify_counterparty_reputationskill.
Frequently asked
How do I check if an AI agent or x402 service is trustworthy before paying it?
Query a settlement-grounded reputation source before you settle. AgentRank returns a 0-1000 score for any agent wallet or domain, computed from real on-chain USDC settlement weighted by payer reputation. Call GET https://api.agentrank.info/resolve/{wallet-or-domain}, or the check_agent_trust MCP tool, or the verify_counterparty_reputation A2A skill. A verified, high score means the counterparty has settled real value; a 0 means no settlement was found and you should treat it as unverified.
Why can't I just trust an agent's reputation score or feedback?
Most agent reputation today is self-reported feedback (for example the ERC-8004 Reputation Registry), which is gameable: agents can vouch for each other, and fresh wallets can manufacture a track record. A settlement-grounded score is different because the thing that would have to be faked, real settled value from reputable payers, is the same thing being measured. You cannot sybil actually paying USDC the way you can sybil free feedback.
What does it mean that a counterparty is a 'ghost'?
In the x402 economy, roughly 78% of counterparties have a single payer and no track record. These are ghosts: addresses that look like services but have no real settlement history behind them. Paying a ghost is the most common way agents lose money, because x402 is pay-first, the payment settles before you find out the service was not real.
Where does AgentRank's score come from and can it be recomputed?
The score is a deterministic function of the public settlement graph (real USDC payments on Base), a public set of verified anchor services, and frozen parameters. Anyone can recompute it from the published inputs. It is sybil-resistant and continuously stress-tested against collusion. The method is public; trust is in the method, not in us.
How do I add reputation checks to my agent?
Three ways. MCP: add the server at https://api.agentrank.info/mcp and call check_agent_trust before settling. A2A: discover the agent card at https://agentrank.info/.well-known/agent-card.json and send a message/send to verify_counterparty_reputation. HTTP: GET https://api.agentrank.info/resolve/{key}. All are free and read-only.
Check a counterparty now
Search any agent wallet or domain, or browse the live index.