Know Your Agent (KYA)

Question 1

What is Know Your Agent (KYA)?

Accepted Answer

Know Your Agent (KYA) is the practice of verifying an AI agent before trusting it with access or money: which agent it is, who built and operates it, who it acts for, and what it is allowed to do. It extends KYC and KYB to software that acts on someone's behalf, and it continues after onboarding.

Question 2

What is a KYA check?

Accepted Answer

A KYA check is the process of verifying an AI agent before it is allowed to transact. It confirms which agent is acting, who built and operates it, which person or business authorized it, and what it may do. Unlike a one-time KYC check, it repeats, because an agent's code, keys and behavior can change.

Question 3

Why is Know Your Agent important?

Accepted Answer

An AI agent can act with someone else's money and access, so a business accepting its requests needs to know which agent it is, who is accountable for it, and whether it still behaves as expected. Without that check, a credentialed agent and an anonymous bot or impersonator look the same, and nobody answers for mistakes.

Question 4

How do you implement a Know Your Agent program?

Accepted Answer

Give every agent a verifiable identity tied to the verified person or business behind it, and record what it may do. Then monitor its transactions and data access as they happen, compare its behavior with a baseline, keep an audit trail, and re-verify whenever its code, keys or permissions change.

Question 5

Who needs Know Your Agent checks?

Accepted Answer

Any business that lets AI agents act on its systems or with its customers' money: merchants and marketplaces accepting agent purchases, banks and payment providers processing agent-initiated payments, and API providers selling access to agents. Companies running their own agents need it too, to know which agents exist and who owns each one.

Question 6

KYA vs KYB: what is the difference?

Accepted Answer

KYB (Know Your Business) verifies a company, including the people who own and control it. KYA verifies an AI agent: which agent it is, what it may do, and which verified person or business it acts for. The two work together, because a company's agents inherit their accountability from the company that passed KYB.

Question 7

What does a KYA provider do?

Accepted Answer

A KYA provider verifies the person or business behind an AI agent, issues the agent a credential that services can check cryptographically, and records what the agent may do. It then watches the agent's activity, so a misbehaving or compromised agent can be flagged and its credentials revoked.

Question 8

What standards exist for Know Your Agent?

Accepted Answer

KYA is still being assembled from several efforts. The IETF Web Bot Auth working group is standardizing how automated clients authenticate to websites. ERC-8004 is a draft Ethereum standard (as of October 2026) for agent identity registries, KYAPay is an individual Internet-Draft for agent identity tokens, and W3C Verifiable Credentials is a Recommendation for tamper-evident credentials.

Question 9

How does Know Your Agent relate to Know Your Intent?

Accepted Answer

Know Your Agent establishes who is acting: which agent, who operates it and who it works for. Know Your Intent checks what that agent may do right now: whether a specific action falls inside what its user authorized. A payment needs both, because a verified agent can still attempt something its user never approved.

Question 10

What is agent onboarding?

Accepted Answer

Agent onboarding is the first KYA step: registering a new AI agent before it may transact. The operator proves who they are, the agent receives a verifiable identity and credentials, its permissions are recorded, and it is linked to the person or business it acts for. Monitoring then continues for as long as the agent runs.

Question 11

What is an agent passport?

Accepted Answer

An agent passport is a signed credential that travels with an AI agent and states who built it, who it acts for and what it may do. Services verify the issuer's signature instead of trusting what the agent says about itself. There is no single standard yet; W3C Verifiable Credentials is one format such a credential can use.

Question 12

How do you score an AI agent's risk?

Accepted Answer

Score the agent on evidence, not claims: whether its identity and operator are verified, how much authority it holds, what it is trying to do, and how its current behavior compares with its own history. Weight unverified operators, high-value or irreversible actions and sudden changes in pattern most heavily, and rescore continuously.

Question 13

How do you verify the developer behind an AI agent?

Accepted Answer

Verify the developer the way you would any counterparty: KYB for a company, KYC for an individual. Then bind the agent to that verified entity cryptographically, for example with a signing key or credential issued after the check, so every request the agent signs traces back to someone accountable.

Question 14

Can an AI agent pass KYC?

Accepted Answer

No. KYC verifies a human's identity, usually with documents and biometrics, and an AI agent has neither. Instead, the person or business behind the agent passes KYC or KYB, and a KYA check links the agent to that verified principal so its actions can be traced back to them.

Question 15

Why is KYC not enough when an AI agent pays?

Accepted Answer

KYC proves who the account holder is, usually once, at onboarding. When an agent pays, the bank sees the account holder's credentials, not which software used them or what the user asked for. KYC cannot tell a legitimate agent from a compromised one, or a requested purchase from one the user never made.

Question 16

What is agent reputation and how is it measured?

Accepted Answer

Agent reputation is the record of how an AI agent has behaved for the people and services it has worked with. It is built from feedback after interactions, independent validation of its work and its transaction history. ERC-8004, a draft Ethereum standard as of October 2026, defines an onchain reputation registry where clients post feedback about an agent.

Question 17

What is an agent trust score?

Accepted Answer

An agent trust score is a single number summarizing how far an AI agent can be trusted right now. It combines whether its identity and operator are verified, its reputation from past interactions and its recent behavior. It speeds up decisions, but it should be recalculated continuously and backed by the evidence behind it.

Question 18

How do you verify which model and version an agent runs?

Accepted Answer

From the outside, usually you cannot; you rely on what the operator declares. Stronger evidence comes from attestation: a signed record of the agent's code, or a trusted execution environment (TEE) attesting to what is running. ERC-8004, a draft Ethereum standard as of October 2026, lists TEE oracles among the methods validators can use to check an agent's work.

Question 19

How do you detect a compromised agent identity?

Accepted Answer

Watch for the agent acting unlike itself: requests from new infrastructure or unfamiliar keys, actions outside its permissions, and spending or data access above its normal pattern. Compare live behavior against its baseline, alert on deviations, and be ready to revoke its credentials quickly when one appears.

Question 20

How often should an AI agent be re-verified?

Accepted Answer

Continuously, with a full re-check on every material change. Monitor each transaction and data request as it happens, and re-verify whenever the agent's code, model, keys, permissions or owner change. A fixed review schedule is a backstop, not a substitute, because an agent can change between reviews.

Question 21

What data should a KYA record store?

Accepted Answer

At minimum: the agent's identifier and public keys, the verified person or business it acts for, its operator or developer, the permissions and limits it was granted, the version of its code or model, when it was verified and by whom, its current status, and an audit trail of its activity and credential changes.