Know Your Agent (KYA): The Identity Standard That Makes Agent Payments Safe
The agent safety question used to center on whether an agent would do something harmful. The payment industry is now asking a harder question: before an agent moves money, can you prove it’s authorized to do so? Know Your Agent (KYA) is the emerging answer — an identity standard that treats agent authorization as a first-class verification problem, not an afterthought.
The short answer
Know Your Agent (KYA) is an identity and authorization framework for AI agents transacting autonomously. Before a payment completes, KYA verifies the agent's identity, the delegation chain from the human or organization that spawned it, and the capability boundaries it is permitted to operate within.
Skyfire coined the term in their Know Your Agent (KYA) framework. It has since been adopted by identity vendors including Sumsub and Trulioo as a standard vocabulary for agent verification. The core insight is simple: payment networks were built assuming a human is ultimately accountable for every transaction. Autonomous agents break that assumption, and KYA rebuilds it.
KYC Was Built for Humans. Agents Need Something Different.
KYC works by tying a real-world identity — a passport, a tax ID, a biometric — to an account. Humans can be verified because they exist continuously, hold government-issued credentials, and are legally liable for their actions.
AI agents are none of those things. An agent is instantiated at runtime, may operate in parallel across thousands of sessions, holds no government document, and cannot be held legally liable in the traditional sense. KYC processes applied naively to agents either fail entirely — an agent has no passport — or produce meaningless results, because verifying the model doesn’t verify the authorization.
Agent credential isolation is a related discipline: giving each agent its own bounded credentials so one compromised agent can’t access another’s resources. KYA goes further. It asks not just whether credentials are isolated, but whether the agent holding those credentials was legitimately authorized to hold them in the first place.
The regulatory environment is moving. Taylor Wessing’s February 2026 analysis of agentic AI in payments identifies authorization chains and transaction logs as the two structural requirements that regulators will use to assess liability when agent payments go wrong. KYA addresses both.
What Is Know Your Agent (KYA)?
KYA is a four-part verification protocol that payment systems can run before authorizing an agent-initiated transaction.
1. Agent Identity — a cryptographically signed credential that identifies this specific agent instance: which model, which version, which deployment, issued by whom. The parallel to a passport is intentional.
2. Authorization Chain — a verifiable record of delegation: the human or organization that controls the agent, the scope of authority granted, and whether that authority extends to the transaction being requested. A procurement agent authorized to buy software licenses should not be able to approve payroll.
3. Capability Boundaries — declared and enforced limits on what the agent can do: spending ceilings per transaction, per day, or per category; allowed counterparties; prohibited action types. Boundaries declared at registration and enforced at the payment layer, not inside agent code.
4. Behavioral Reputation — consistency of prior actions against declared intent. An agent that has never exceeded a $200 transaction limit suddenly requesting $5,000 warrants a hold, regardless of valid credentials.
Sumsub’s know-your-agent writeup notes that the fourth pillar — behavioral reputation — is where most current implementations are weakest. Credentials and delegation chains are solvable with cryptography. Reputation requires longitudinal data, which is why wallet tenure and transaction history matter.
The Four KYA Questions Every Payment System Should Answer
Trust in autonomous agents is a practical question, not a philosophical one. KYA operationalizes it into four specific checks that a payment processor, API gateway, or orchestration layer can run in real time.
| KYA Question | What It Checks | How It’s Verified |
|---|---|---|
| Who is this agent? | Agent identity, model, version, deployment | Signed identity credential or Digital Agent Passport |
| Who authorized it? | Delegation chain from principal to agent | Cryptographic delegation record |
| What is it allowed to do? | Capability boundaries, spend limits, allowed merchants | Policy manifest issued at agent registration |
| Has it behaved consistently? | Transaction history vs. declared intent | Wallet tenure, prior transaction records |
A payment that cannot answer all four questions is, by KYA standards, unauthorized — regardless of whether the credentials themselves are valid. A stolen credential chain with no transaction history raises the same flag as an agent with no credentials at all.
The Trulioo KYA whitepaper frames this clearly: identity without authority is insufficient, and authority without boundary is dangerous. The framework requires all three to coexist before authorization is granted.
How KYA Connects to Spending Limits and Verifiable Intent
Mastercard’s concept of Verifiable Intent is a practical implementation of KYA’s third pillar — capability boundaries — expressed at the payment network level. Before Mastercard Agent Pay processes a transaction, the agent must present verifiable evidence that the action falls within the scope for which it was authorized. A shopping agent instructed to find the cheapest hotel under $150 per night cannot, under Verifiable Intent, book a $400 penthouse without a new authorization event.
This architecture matters because it shifts the trust burden from the agent to the infrastructure. An agent claiming it was authorized to do something is not verification. A signed authorization record issued before the action started, tied to a declared spending limit, is.
The table below maps KYA pillars to practical payment controls:
| KYA Pillar | Practical Control | Example |
|---|---|---|
| Identity | Signed agent credential | Agent ID tied to issuing organization |
| Authorization chain | Delegation record with scope | ”Authorized for SaaS tools, max $500/month” |
| Capability boundaries | Hard spending limits | Per-transaction cap enforced at infrastructure |
| Behavioral reputation | Anomaly detection | First $5k transaction flagged for human review |
Spending limits are the most tractable control. They can be set structurally — not as promises in agent code — and enforced at the payment layer before any money moves. KYA without spending limits is an identity framework that still allows unlimited damage. The combination is what makes agent payments safe.
Practical KYA: What to Check Before Authorizing an Agent Transaction
Implementing a financial zero-trust model for agents means running KYA checks at authorization time, not after the fact. Here is a practical checklist for any team deploying agents that spend money.
Before deploying the agent:
- Issue a signed identity credential for the agent instance, tied to the deploying organization
- Define the delegation scope in writing: what the agent can authorize, for how much, with which counterparties
- Set hard spending limits at the payment infrastructure level — not in agent code
At transaction time:
- Verify the agent credential is current and has not been revoked
- Confirm the requested transaction falls within the declared capability boundaries
- Check behavioral consistency: is this transaction anomalous relative to prior history?
- Log the full authorization chain — agent ID, delegation record, transaction amount, merchant, timestamp — before the payment clears
Post-transaction:
- Maintain an immutable record of every transaction against the authorization chain
- Flag any transaction that was authorized outside normal parameters for human review
- Use transaction history to build behavioral reputation for future checks
The Taylor Wessing analysis makes clear that liability in agentic payments will follow the authorization chain. If an agent spends beyond its declared limits and no one can produce the authorization record, the party that deployed the agent bears the full liability. KYA documentation is not compliance theater — it is the legal record that determines who is responsible when something goes wrong.
ATXP credits are issued to agent accounts with hard spending limits and transaction-level receipts. Every payment carries an authorization record: which agent, under which principal, against which declared limit. That is KYA-compatible infrastructure by design — not a feature added after the fact.
FAQ
What is Know Your Agent (KYA)? Know Your Agent (KYA) is an identity verification standard for autonomous AI agents. Before a payment is authorized, KYA verifies the agent’s identity, the delegation chain from the controlling human or organization, the capability limits the agent was given, and whether its behavior is consistent with prior actions.
How is KYA different from KYC? KYC verifies human identity using government documents and biometrics. KYA verifies agent identity using cryptographic credentials, signed authorization chains, and declared capability boundaries. Humans have passports; agents have signed identity manifests and policy-bound spending limits.
Is KYA required by regulation? Not yet as of March 2026. KYA is an emerging industry standard. However, the EU AI Act’s enforcement mechanisms beginning August 2026, combined with regulatory analysis from Taylor Wessing (February 2026), strongly suggest that agent identity and transaction log requirements will be formalized. Building KYA-compliant infrastructure now is lower cost than retrofitting it under a deadline.
Which companies have adopted the KYA framework? Skyfire coined KYA. Identity verification vendors Sumsub and Trulioo have both published KYA-specific frameworks. Mastercard’s Verifiable Intent concept operationalizes KYA’s capability boundary pillar at the payment network level. The knowyouragent.network project tracks industry adoption.
What happens if an agent transacts without KYA verification? The transaction may complete, but liability is unresolved. Taylor Wessing’s February 2026 analysis notes that when agent payments go wrong without authorization records, liability defaults to the deploying organization. Without a KYA record — identity, authorization chain, capability limits — there is no documentation to limit that liability.