Let AI agents pay — without handing them the keys.
AgentTrust is the compliance and authorization layer for AI-agent payments. An agent can request a payment; it never holds uncontrolled authority to move money. Deterministic services decide — fail-closed.
The core principle
An AI agent can request a payment. It must never have uncontrolled authority to move money. LLM output never executes a financial action — deterministic services decide, and the system is fail-closed throughout.
From agent request to signed authorization.
Every agent payment is composed through an ordered pipeline. No stage can be skipped, and only an approved request ever receives authority to move money.
Identity & lifecycle (KYA)
Every agent, provider and principal is registered with a lifecycle status and Know-Your-Agent evidence, classified for risk before it can act.
Delegated authority
Agents only hold the scoped capabilities and delegations you grant. Each request checks authority — approve, require a human, or deny.
Intent verification & match
The agent's intent is cryptographically verified and matched to the actual payment. A mismatch is caught and stopped.
Compliance screening
The request runs through your existing AML, sanctions, fraud and beneficiary checks — with the agent's context injected.
Trust, risk & behaviour
A trust score plus behaviour baselines flag anomalies, account takeover and unexpected tool- or model-changes.
Decision
One explainable decision composes everything above: approve, step-up, send for human approval, hold, reject or block.
Human approval
Anything high-risk lands in a human-in-the-loop queue with the full evaluation context attached.
Scoped signed authorization
Only an approved request gets a short-lived, single-use signed authorization — never a standing key to move money.
Payment orchestration
The valid authorization is routed to your payment rail through pluggable network adapters.
One decision, six honest outcomes.
The decision engine never quietly moves money. It returns one of six outcomes, each with the reasons behind it.
Active agents
128
Auto-approved
94%
Human / step-up
5%
Blocked / rejected
1%
Recent decisions
| shopper-agent-01 | $1,240 | APPROVE |
| procure-bot | $18,500 | HUMAN_APPROVAL |
| travel-agent-eu | $430 | APPROVE |
| ops-agent-07 | $9,900 | STEP_UP |
| unknown-agent | $62,000 | BLOCK |
Everything an agent needs — and nothing it doesn't.
Agent Registry & KYA
Register agent providers, principals and agents with deterministic Know-Your-Agent evidence, risk classification and full lifecycle.
Delegated Authority
Grant scoped capabilities and delegations; every request is checked (approve / human-approval / deny).
Verifiable Intent
Verify the agent's intent and match it to the payment — with a boundary scanner that catches intent–payment mismatches.
Policy Engine
Versioned, JSON condition-tree policies you can simulate against scenarios before they ever go live.
Trust & Behaviour
Trust scoring, behaviour baselines and anomaly, takeover and tool/model-change detection.
Scoped Authorizations
Issue short-lived, single-use signed authorizations; validate and revoke them centrally.
Human-in-the-Loop
Route high-risk requests to an approval queue with the complete decision context.
Hash-Chained Audit
Every decision is written to a SHA-256 hash-chained, append-only ledger you can prove wasn't altered.
Isolated, auditable, reversible.
Ready for the agent economy?
See AgentTrust evaluate a real agent payment end to end — request, decision, authorization and audit.