KYC (Know Your Customer) and KYB (Know Your Business) are due-diligence duties that identify and verify the human or legal entity you do business with, largely for anti-money-laundering and counter-terrorist-financing purposes. KYA (Know Your Agent) verifies the autonomous software agent acting for that person or company: what it is, who is accountable for it, what it is allowed to do, and whether it can be revoked.
Short version: KYC/KYB verify the principal. KYA verifies the agent.
They are not rivals, and this is the most common thing teams get wrong. KYC, KYB and KYA stack. Customer due diligence tells you who the accountable person or company is; KYA tells you which software is acting for them, under what authority, and how to stop it. You rarely do one instead of another. For the full picture of the agent-side discipline, see What is Know Your Agent (KYA)?.
This page sets out what each term actually covers, where they overlap, a worked example, the mistakes that follow from conflating them, and where regulation and standards touch each.
What KYC actually is
KYC is not one law; it is the industry name for a set of customer due diligence (CDD) obligations countries impose on financial institutions and other regulated firms.
The FATF Recommendations set the global baseline. Recommendation 10 requires firms to “undertake customer due diligence (CDD) measures” and prohibits keeping anonymous accounts or accounts in obviously fictitious names. In the United States, the bank customer identification rule (31 CFR 1020.220) requires risk-based procedures for verifying “the identity of each customer,” to the extent reasonable and practicable, so that the bank can “form a reasonable belief that it knows the true identity of each customer.”
KYC answers: Is this customer who they say they are, and what is their risk?
What KYB adds
KYB is the industry label for applying due diligence to a legal entity rather than a natural person. It verifies the company’s registration, its status, and — critically — its beneficial ownership: the people who ultimately own or control it.
The underlying duties come from the same AML regime. FATF’s beneficial-ownership work centres on the principle that “only a natural person can be an ultimate beneficial owner, and more than one natural person can be the ultimate beneficial owner.” In the United States, the beneficial ownership rule (31 CFR 1010.230) requires covered financial institutions to identify each individual who owns “25 percent or more of the equity interests” of a legal entity customer, plus “a single individual with significant responsibility to control, manage, or direct” it.
KYB answers: Is this business legitimate, and who really controls it?
What KYA adds
KYA keeps the discipline of “identify and verify before you trust” but shifts the subject to the agent and the question to “what may it do right now”.
An agent is a new identity category, not a rename of a service account. As the Cloud Security Alliance puts it, agents “authenticate programmatically, execute instructions at machine speed, operate across trust boundaries that span organizational perimeters, spawn child agents dynamically, and persist long after the business context that originally justified their privileges has changed.” Microsoft’s agent platform describes the same shift, adding that organisations need “purpose-built identity constructs to authenticate, authorize, govern, and protect these nonhuman identities.”
That produces five dimensions KYC was never built to express:
- Delegation. The agent acts on behalf of a principal. That authority must be explicit, scoped, and provable — the subject of new IETF OAuth work for AI agents.
- Per-action permission. A verified agent is not thereby allowed to do anything. KYA bounds what it may do, at what limits, for how long.
- Machine-verifiable identity. Rather than documents reviewed by a human, agent identity leans on cryptographically verifiable credentials and attested workload identity, such as SPIFFE’s short-lived identity documents.
- Fast revocation. An agent can be compromised or misaligned in seconds. Authority must be withdrawable at the same speed.
- Auditability. High-stakes regimes already demand event logs; KYA extends that expectation to agent actions and the authority behind them.
Side-by-side: KYC, KYB and KYA
| KYC (Know Your Customer) | KYB (Know Your Business) | KYA (Know Your Agent) | |
|---|---|---|---|
| Subject verified | A natural person | A legal entity and its beneficial owners | An autonomous software or AI agent |
| Core question | Who is this customer? | Who really owns or controls this company? | Who does this agent act for, what may it do, how is it revoked? |
| Origin | AML/CFT regulation (FATF Recommendation 10) | AML/CFT regulation (beneficial ownership, FATF Recommendation 24) | Emerging standards and vendor practice (NIST NCCoE, OpenID Foundation, cloud IdPs) |
| Typical evidence | Government ID; name, date of birth, address | Entity documents and beneficial-owner identification (“25 percent or more” plus a controller) | Cryptographic identity and attestation; delegation; scoped tokens |
| Reuses | — | KYC data about owners | Existing identity standards (OAuth, SPIFFE, MCP authorisation) |
| Authority model | Relationship-level, reviewed periodically | Relationship-level | Per-action, scoped, time-bound, continuously evaluated |
| Lifecycle | Onboarding plus periodic review | Onboarding plus periodic review | Enrol, verify, authorise, monitor, revoke — continuously |
| Who is accountable | The customer | The entity’s controlling persons | The agent’s principal (a person or organisation) |
| Primary driver | AML/CFT regulation | AML/CFT regulation | Access control, fraud, accountability, agent governance |
Where the three overlap
The overlap is real, and it matters for compliance teams:
- The principal is still a KYC/KYB subject. To do KYA you must know who the principal is — which is exactly the KYC/KYB question. KYA does not replace customer due diligence; it depends on it.
- Agent misuse can be a financial-crime risk. An unverified agent acting for an anonymous principal is an AML/CDD problem, not just a security one.
- Both need assurance levels. NIST’s Digital Identity Guidelines define assurance levels for people and their authenticators (SP 800-63-4); KYA needs the analogous notion for agents and the software they run.
- Governance frameworks cover both. AI governance and the existing AML programme both ask an organisation to know and control what acts on its behalf; agent identity is part of the AI-governance picture.
Worked example: a verified company’s procurement agent pays an invoice
This scenario is illustrative — it shows how the layers combine, not a description of one product.
A supplier’s finance team lets an AI agent pay invoices on the company’s behalf.
- KYB on the buyer. The supplier verifies the buying company as a legal entity and identifies the individuals who own or control it, using the same beneficial-ownership discipline required for legal-entity customers.
- KYC on the human who authorised the agent. The specific person who set up the mandate is identified and verified as the accountable human behind it.
- KYA on the agent. The agent is enrolled with a verifiable cryptographic identity bound to the company as principal. It is not identified by a shared API key.
- Scoped mandate. The agent holds a delegation granting it authority to pay invoices, up to a limit, until an expiry date — authority that is explicit and provable rather than implied.
- Per-payment check. Each payment is evaluated against current policy. An invoice above the limit triggers a step-up approval from the human; the invoice is rejected.
- Record and revoke. Every payment logs the agent, the principal, the mandate and the time. If the finance team revokes the mandate, the next payment fails.
Steps 1–2 are KYC/KYB. Steps 3–6 are KYA. Neither alone produces accountability for the payment.
How they fit together in practice
A practical sequence for a service that accepts agents:
- KYB/KYC the principal — establish the accountable person or entity (existing duties).
- Verify the agent — establish a verifiable agent identity, tied to the principal.
- Bind authority — issue a scoped, time-bound, revocable credential.
- Check per action — is this request within the agent’s current authority?
- Log and revoke — record what happened; be able to cut the agent off.
Step 1 is KYC/KYB. Steps 2–5 are KYA. Together they form one chain of accountability from a verified customer to a specific agent action. For the verification mechanics, see how to verify an agent; for protocol-level examples, see code examples.
Common mistakes when teams conflate KYC and KYA
- Thinking KYA replaces KYC/KYB. It does not. The principal still needs customer due diligence where regulated activity is involved. KYA is an additional layer, not a substitute.
- Treating the agent’s account as the customer. An agent acting for a company is not itself the customer; the company (or person) remains the accountable principal.
- Applying relationship-level checks to per-action risk. KYC establishes a relationship once and reviews it periodically. An agent’s authority must be evaluated per action, because an agent’s behaviour is generated and can vary between requests.
- Reusing customer credentials for the agent. Handing an agent a customer’s own credentials gives it the customer’s full access and cannot be revoked narrowly. Delegation should be explicit and scoped, not credential-sharing.
- Assuming “say it’s an AI” equals knowing the agent. The EU AI Act’s transparency duty means people must be told they are interacting with an AI system. That is a disclosure obligation, not identity verification — it does not tell you whose agent it is or what it may do.
- Ignoring offboarding. Non-human identities that are not properly deactivated remain exploitable. Agents drift out of their original business context; an inventory and lifecycle process is what catches that.
Where regulation and standards touch each
| Layer | What applies | Type |
|---|---|---|
| KYC | FATF Recommendation 10; bank customer identification rules (e.g. 31 CFR 1020.220) | Binding AML/CFT regulation |
| KYB | FATF beneficial-ownership standards (Recommendation 24); legal-entity beneficial-ownership rules (e.g. 31 CFR 1010.230) | Binding AML/CFT regulation |
| AI transparency | EU AI Act Article 50 — inform people they are interacting with an AI system, from 2 August 2026 | Binding regulation (transparency only) |
| Agent identity and authorization | NIST NCCoE project on software and AI agent identity and authorization | Standards work in progress |
| Delegation semantics | OAuth 2.0 (RFC 6749); OAuth token exchange (RFC 8693); IETF on-behalf-of work for AI agents | Open specifications |
| Agent tool access | Model Context Protocol authorisation, built on OAuth | Open specification |
| Agent identities in practice | Cloud identity providers shipping dedicated agent identities (Microsoft Entra Agent ID; Google Cloud Agent Identity) | Vendor-documented |
The pattern is consistent: the customer side is regulated and mature; the agent side is standards-led and still converging. No single regulation currently requires full “KYA”, and no single standard defines it end to end.
FAQ
Does KYA replace KYC? No. It adds a layer. You still perform customer due diligence on the principal; you now also verify and bound the agent.
Is KYB different from KYC? KYB is the business-entity case of the same due-diligence idea. “KYB” is an industry term for corporate CDD and beneficial-ownership checks; it is not a formally defined FATF term.
Which regulation requires KYA? None names it. Specific obligations apply — AML customer due diligence, and the EU AI Act’s transparency duty for AI systems from 2 August 2026 — and KYA is the practice that connects identity, authority and accountability across them. It is not itself a named legal obligation.
Is KYA a vendor category? Partly. Vendors cover pieces (workload identity, verifiable credentials, authorization, agent security). Few cover the whole chain; check which of the five questions each actually answers.
Can one agent serve several principals? It can, but each authority must be bound to the correct principal. Multi-principal agents make the delegation link the critical control — the system must know, per action, whose authority is being exercised.
This is an explanation, not legal advice.
Keep reading
- What is Know Your Agent (KYA)?
- Why agent identity is a new problem
- How to verify an agent
- Code examples
- Glossary: principal, delegation, agent identity, audit trail
Sources
- FATF, The FATF Recommendations (Recommendation 10, Customer Due Diligence) — https://www.fatf-gafi.org/en/publications/Fatfrecommendations/Fatf-recommendations.html (accessed 2026-10-03)
- FinCEN / U.S. Federal Register, Customer Due Diligence Requirements for Financial Institutions (81 FR 29398) — https://www.federalregister.gov/documents/2016/05/11/2016-10567/customer-due-diligence-requirements-for-financial-institutions (accessed 2026-10-03)
- U.S. eCFR, 31 CFR § 1020.220 — Customer identification program requirements for banks — https://www.ecfr.gov/current/title-31/section-1020.220 (accessed 2026-10-03)
- FATF, Guidance on Beneficial Ownership of Legal Persons (March 2023) — https://www.fatf-gafi.org/content/dam/fatf-gafi/guidance/Guidance-Beneficial-Ownership-Legal-Persons.pdf.coredownload.pdf (accessed 2026-10-03)
- U.S. eCFR, 31 CFR § 1010.230 — Beneficial ownership requirements for legal entity customers — https://www.ecfr.gov/current/title-31/subtitle-B/chapter-X/part-1010/subpart-B/section-1010.230 (accessed 2026-10-03)
- NIST NCCoE, Agentic AI Identity and Authorization Project Resource Hub — https://pages.nist.gov/nccoe-ai-identity/ (accessed 2026-10-03)
- OpenID Foundation, Identity Management for Agentic AI (whitepaper, October 2025) — https://openid.net/wp-content/uploads/2025/10/Identity-Management-for-Agentic-AI.pdf (accessed 2026-10-03)
- Cloud Security Alliance, Agent Identity Governance Framework v1 — https://labs.cloudsecurityalliance.org/agentic/agentic-identity-governance-framework-v1/ (accessed 2026-10-03)
- Microsoft Learn, What is Microsoft Entra Agent ID? — https://learn.microsoft.com/en-us/entra/agent-id/what-is-microsoft-entra-agent-id (accessed 2026-10-03)
- Google Cloud, Agent Identity overview — https://docs.cloud.google.com/iam/docs/agent-identity-overview (accessed 2026-10-03)
- SPIFFE, SPIFFE Concepts — https://spiffe.io/docs/latest/spiffe-about/spiffe-concepts/ (accessed 2026-10-03)
- IETF, RFC 6749: The OAuth 2.0 Authorization Framework — https://www.rfc-editor.org/rfc/rfc6749.html (accessed 2026-10-03)
- IETF, RFC 8693: OAuth 2.0 Token Exchange — https://www.rfc-editor.org/rfc/rfc8693.html (accessed 2026-10-03)
- IETF, draft-oauth-ai-agents-on-behalf-of-user-02 — https://datatracker.ietf.org/doc/html/draft-oauth-ai-agents-on-behalf-of-user-02 (accessed 2026-10-03)
- Model Context Protocol, Authorization — https://modelcontextprotocol.io/specification/draft/basic/authorization (accessed 2026-10-03)
- EU AI Act, Article 50: Transparency Obligations for Providers and Deployers of Certain AI Systems — https://artificialintelligenceact.eu/article/50/ (accessed 2026-10-03); official text: Regulation (EU) 2024/1689
- European Commission, Guidelines on transparency obligations for providers and deployers of certain AI systems — https://digital-strategy.ec.europa.eu/en/policies/guidelines-ai-transparency-obligations (accessed 2026-10-03)
- OWASP, Non-Human Identities Top 10 — https://owasp.org/projects/non-human-identities-top-10 (accessed 2026-10-03)
- NIST, SP 800-63-4: Digital Identity Guidelines — https://pages.nist.gov/800-63-4/ (accessed 2026-10-03)
Know Your Agent (KYA) explains agent identity, verification and accountability. This is an explainer, not legal or compliance advice — see our Sources & methodology. New to KYA? Start with What is Know Your Agent? and the glossary.
Last reviewed: 8 October 2026. Sources are listed with the page and in our Sources & methodology.