The answer depends entirely on how the agent paid. Pay-then-serve protocols (x402 and similar) are designed so the agent pays first and the service responds; if the service does not respond, the payment is still final, because that is the point of the design. It is fine for a cent per API call. It is the whole loss for a purchase, a job or a trade. The question to ask before any agent pays is where the money is while the other side delivers.
What happens when the agent used escrow
- 1On iEscrow the agent did not pay anyone. It opened a trade; the counterparty’s crypto (or yours, for a purchase) was locked in escrow, and the funding step was approved by you, by link with your second factor.
- 2If the other side does not deliver, the trade does not release on its own. For a crypto-for-fiat trade the counterparty confirms your payment or the trade escalates to a dispute; for a product or service you confirm delivery or open a dispute.
- 3A human arbiter decides the dispute with the evidence both sides submit, and writes down the reason. The agent can watch the state of the trade (wait_for_trade_update) and tell you; it cannot decide anything.
- 4If delivery never happens and the window expires, the trade resolves by its rules: funds return to whoever put them in, or the dispute is decided. Either way the money was never in the hands of someone who could walk away with it.
What escrow does not fix
- A purchase your agent made through a different rail. If it paid an address directly, no escrow was involved and the usual rule applies: crypto sent is crypto gone.
- Quality. If what arrived is not what the agent described in the trade, the arbiter reads that description. The precision of what the agent wrote is the protection you get.
The practical test for any agent payment tool: "If the other side disappears after my agent pays, who has the money?" If the answer is "the other side", it is not escrow.