Know Your Agent (KYA) compliance in Europe: EU AI Act, PSD3 and agent-payment liability

What is Know Your Agent (KYA)?

Know Your Agent (KYA) is a four-part verification that a payment system runs before authorizing an agent-initiated transaction: identity, delegation chain, capability boundaries and audit trail. It extends the familiar Know Your Customer (KYC) and Know Your Business (KYB) regimes to a new class of actor — autonomous software that can open new wallets at zero cost, act without a human in the loop, and scale activity horizontally faster than any human attacker.

The core shift is the question being asked. KYC asks "who is this person?" KYA asks "what is the operational reputation and authority of this agent?" — is it sanctions-clean, does it have a verifiable identity, did a real principal authorize it, and is it staying inside the limits it was granted.

Why does KYA matter now? The 2 August 2026 deadline

The forcing function is regulatory. The EU AI Act's obligations for high-risk AI systems take full effect on 2 August 2026. The law mandates human oversight for high-risk systems, which in payments effectively requires a human-binding approach to agent authority — exactly what a delegation chain provides. At the same time, US standards bodies are moving: NIST's AI Agent Standards Initiative published a concept paper on agent identity and authorization in March 2026, proposing to adapt OAuth, OpenID Connect and SPIFFE for agents.

The legal analysis is converging on two structural requirements. As Taylor Wessing's February 2026 review of agentic AI in payments noted, regulators will assess liability when agent payments go wrong using authorization chains and transaction logs. Without a KYA record — identity, authorization chain, capability limits — liability defaults to the deploying organization. Building KYA-compliant infrastructure now is materially cheaper than retrofitting it under a deadline.

What is the difference between KYC, KYB and KYA?

Which EU regulations drive KYA?

No single rule says "do KYA," but four regimes each require a piece of it. Together they make KYA the practical way to comply.

• EU AI Act — high-risk obligations from 2 August 2026. Article 12 requires automatic, tamper-resistant logging over the system lifecycle; Article 14 requires human oversight; Article 50 sets transparency duties. Annex III classifies AI used in creditworthiness, fraud detection, AML risk profiling and automated financial decisioning as high-risk. (artificialintelligenceact.eu)
• PSD2 → PSD3 — PSD2 Strong Customer Authentication assumes a human authorizes each payment; it does not mention AI. PSD3 is expected to start formalizing delegated payment initiation by agents and Strong Customer Authentication for autonomous actors. Until then, fiat-side agent payments still need a human authorization anchor — which a delegation chain records.
• DORA — the Digital Operational Resilience Act requires financial entities to manage third-party (including AI) operational risk, which means knowing and governing the agents in your flows.
• MiCA Travel Rule — for crypto transfers, originator and beneficiary information must travel with the transaction above threshold. For agent stablecoin payments, that requires identifying the agent and its operator. AsterPay automates the Travel Rule for transfers over EUR 1,000.

What are the four checks in a KYA verification?

A KYA system runs four checks before a payment is authorized. Each maps to a regulatory requirement above.

1. Identity — the agent has a verifiable identity, ideally on-chain (e.g. ERC-8004) and bound to its operator. This satisfies the "who acted" question.
2. Delegation chain — cryptographic proof that the human or organization authorized this specific agent to act on its behalf, within stated scope. This is the authorization-chain requirement regulators will test.
3. Capability boundaries — declared and enforced limits: spend per transaction, per day, per category; allowed counterparties; prohibited actions. Crucially, these are enforced at the payment layer, not inside agent code, so a compromised or misaligned agent cannot exceed them.
4. Audit trail — a tamper-resistant log of identity, authorization and the decision, sufficient to attribute liability if a payment is later disputed. This is the Article 12 logging requirement.

Inbound KYA vs internal KYA

KYA splits into two deployments, and conflating them is a common mistake.

• Inbound KYA — governing external customer agents that pay you or act on accounts you hold. You verify the agent before it transacts. This is largely crypto-native or in pilot today, because fiat-side Strong Customer Authentication still expects a human.
• Internal KYA — governing the AI agents your own organization deploys in fraud, KYC, AML, onboarding and financial decisioning. These are the high-risk systems the EU AI Act regulates directly, so the need here is immediate, not future.

AsterPay's KYA is primarily an inbound control: it verifies the paying agent on behalf of the merchant, so the merchant inherits a compliant record without building agent governance themselves.

How does AsterPay implement KYA?

AsterPay runs KYA on every agent payment it facilitates. The model is an open 0-100 trust score across seven components, mapped to five payment tiers with enforced limits:

Sanctions screening runs on every transaction, the Travel Rule is automated for transfers over EUR 1,000, and the seven-component breakdown doubles as the explanation required under the EU AI Act's transparency duties. The KYA v1 schema is published openly at /.well-known/kya-schema-v1.md so other facilitators and merchants can adopt it as a shared trust language — the way FICO became a default credit-score language without being owned by a single bank. The full scoring rubric and tier limits are in What is KYA (Know Your Agent)?

How do I check an agent's KYA score?

A free, unauthenticated GET request returns the score, tier, component breakdown, sanctions status and recommended limits:

curl https://x402.asterpay.io/v1/agent/trust-score/0xYourAgent

{
  "score": 72,
  "tier": "trusted",
  "blocked": false,
  "sanctions": { "clean": true, "provider": "chainalysis" },
  "limits": { "maxPerTx": 5000, "maxDaily": 25000 }
}

Related reading

• What is KYA (Know Your Agent)? — the scoring framework
• ERC-8004 explained — on-chain agent identity
• How do AI agents pay merchants?
• What is x402 EUR settlement?
• MiCA-aligned stablecoin settlement

Have a correction or want this page to cover a different angle? Email [email protected]. We update this page every quarter.