Automation scales transactions and services. But also the risk of failure, and the need
for trust.
1 Human intent
→
∞ Agent actions
→
PaymentsServicesFailuresDisputes
Agents will not need less governance because they are automated.
They will need governance that operates at their speed.
From first contact to final resolution.
Kleros secures every stage of the transaction.
Stage 01 / 04 · Verify
Trust unlocks permissions.
First, find the right agent for the job. Then decide what it may do. Openly verifiable
rules filter out review farming, scams, and badges with unclear provenance before an
agent gets a card, an API key, private data, or clearance for a sensitive task.
Prove expertise, identity, language, or jurisdiction.
An agent can pay for data that never arrives. Escrow holds value until work is done.
Arbitration and recovery stop a failed machine payment from becoming a permanent loss.
Pay after proof — release funds to vendor only on delivery.
Pay before proof — if failed, vendor will compensate customer.
An experimental agent lost 0.9 USD after a failed x402 payment. Would it be sustainable to involve humans every time to asses who's fault?
Technical paths +
x402MPPERC-8183EscrowArbitration
Without the return path, a failed machine payment is simply gone.
Stage 03 / 04 · Optimize
Automate the clear. Escalate the uncertain.
Cheap, obvious cases should not wait for an expensive process. Complex, sensitive, or
high-value cases can move to specialized agents and then to human jurors.
AI can triage before a court and help jurors process evidence inside it. Humans keep the authority to reverse.
40%20%40%
Clear claimant · uncertain · clear respondent. Illustrative hypothesis, not a forecast.
Cheap, obvious cases settle themselves. Hard ones buy more computing power, or a human.
Will a person feel confident enough to reverse an AI?
Human escalation only matters if it carries real authority. Kleros CEO Federico Ast
examines the tension between machine confidence and meaningful human oversight.