Know Your Agent (KYA): identity verification for AI agents
Know Your Agent (KYA) is the practice of verifying an AI agent before you let it act: who operates it, on whose behalf it acts, which limits it was given, and how it has behaved so far. It applies the logic of Know Your Customer (KYC) to software that buys, books, trades and calls APIs without a human in the loop.
A KYA check answers four questions in a form a machine can read and a third party can recompute:
- Identity — who created and operates this agent.
- Authorization — whose instructions it executes and which limits apply: amounts, counterparties, actions, expiry.
- Behavior history — whether it has stayed within its stated parameters.
- Portability — whether any counterparty can verify all of the above independently, without asking the issuer.
KYA and KYC
KYC verifies a person once, at onboarding, against documents and a regulator’s rulebook. KYA verifies software, often on every request, against a signed record of who stands behind it and what it may do.
| KYC | KYA | |
|---|---|---|
| Subject | a person or company | an AI agent acting for a person or company |
| Evidence | passport, register extract, proof of address | decentralized identifier, signed credentials, mandate |
| When | at onboarding, periodic review | before each consequential action |
| Checked by | a compliance team | the counterparty’s software, automatically |
| Key question | is this person who they claim to be | is this agent allowed to do this, for whom, up to which limit |
KYA does not replace KYC. The principal behind an agent may still need KYC; KYA links the agent to that principal and to the limits the principal set.
Why KYA matters now
In financial services, non-human identities already outnumber human employees 96 to one (Sean Neville, a16z crypto, January 2026). Merchants, APIs and marketplaces that cannot tell an authorized agent from an unknown script have two options today: block agents at the firewall, or accept them without knowing who is liable. KYA gives them a third option: accept the agent, check its credentials, and enforce its limits.
The cases where this is already operational are those where an agent moves money or makes commitments: shopping agents, trading and payment agents, agents bidding in ad auctions, and agents placing positions on prediction markets.
How a KYA check works
A counterparty receives a request from an agent that presents a decentralized identifier (DID, a globally unique identifier the agent controls with its own key). The counterparty then:
- Resolves the identifier to a DID document with the agent’s public keys and registration metadata.
- Verifies the credentials: W3C Verifiable Credentials, signed statements about the agent that anyone can check against the issuer’s key.
- Checks the mandate: the signed authorization from the principal, with spending limits, permitted actions and an expiry, and whether this request falls inside it.
- Looks at the track record and decides: allow, deny, or allow with conditions.
With MolTrust, the first steps are public HTTP calls, no account and no API key required:
# 1. Resolve the agent's DID document
curl https://api.moltrust.ch/identity/resolve/did:moltrust:<id>
# 2. Verify its registration and credentials
curl https://api.moltrust.ch/identity/verify/did:moltrust:<id>
# 3. Get an allow/deny decision against a minimum trust score
curl https://api.moltrust.ch/trust/gate/did:moltrust:<id>
The gate denies by default: an unknown agent or an agent whose score is withheld gets DENY with a reason, never a silent pass.
How MolTrust implements KYA
- Identity: every agent gets a
did:moltrustidentifier, resolvable by anyone, with its keys and registration recorded in the DID document and the registration transaction on Base. - Authorization: the Agent Authorization Envelope (AAE), a signed mandate that binds an agent to its principal, its permitted actions and its spending limits. The AAE is published as an IETF Internet-Draft.
- Enforcement: each transaction is checked against the mandate before it executes, and the verdict can be recomputed by the counterparty from the same inputs.
- Evidence: credentials are anchored on Base, an Ethereum layer-2 network, so their existence at a given time can be proven without trusting MolTrust (how anchoring works).
Frequently asked questions
Is KYA a legal requirement?
Not as a named obligation. Regulators in the EU, the UK and the US are examining how liability works when software acts on someone’s behalf, and KYA is how a business documents who authorized an agent and within which limits.
Does KYA identify the person behind an agent?
It links the agent to its principal. How much of the principal’s identity is disclosed depends on the credential: a counterparty may only need proof that a verified principal exists and that the amount is within the mandate.
How is KYA different from API keys or OAuth?
An API key proves that a caller holds a secret. OAuth delegates access to a resource. KYA adds who the agent acts for, what it may spend or commit, and a record a third party can verify without contacting the issuer.
Can an agent pass KYA and still misbehave?
Yes. KYA establishes accountability and limits; it does not predict behavior. Mandate enforcement stops actions outside the limits, and the track record makes past behavior visible to the next counterparty.
Further reading
- KYA Whitepaper — the strategic framework behind the four pillars
- Budget limits for AI agents
- Who proves what the agent was allowed to buy?
- Developer documentation — register an agent and integrate verification