KYC verifies people.
Agents need KYA.
AI agents already move real money — payments, subscriptions, trading, procurement. Yet before one pays — at the open, any-agent-callable layer — few can answer all three: who is it? can it prove it's really itself? and if something goes wrong, can you revoke it in one click? OceanAlt turns those three into a single gate the agent can't switch off, enforced before settlement. That's KYA — Know-Your-Agent.
Three questions, three gates
① Attribution
KYAWhich accountable entity is this agent attributed to?
Without it: An anonymous agent spends freely, and when something breaks there's no one to hold accountable.
We: At registration the agent is bound to an accountable entity; any unattributed payment is blocked outright.
② Proof
KYA/ProofCan the caller prove “I am this agent”?
Without it: Anyone who knows your agentId could spend against your budget. This is exactly where we broke our own gateway in Attack Lab #3.
We: Registration issues a one-time credential; every later call must present it. Impersonation with no credential → blocked.
③ Revocation
KYA/RevokedIf the credential leaks or the agent goes rogue, can you kill it instantly?
Without it: One leak = bleeding forever.
We: One-click revoke, effective immediately; every payment from that agent dies on the spot afterward.
Walk through a KYA lifecycle yourself
Don't take our word for it — watch: the agent below gets an ID card, proves itself to pay, gets blocked when impersonated, is revoked in one click, and then — even with a still-valid credential in hand — can't pay after revocation. Every step is a real call to the live gateway.
Hit the button to watch an agent's “life”: gets an ID → proves itself → impersonation blocked → revoked in one click → blocked even with a valid credential after revocation. All real gateway calls.
KYA doesn't stop at identity — it's gate one of a full firewall
KYA answers “who.” Behind it, more gates answer “should this payment happen, and can it”: per-tx limit, daily velocity, payee allowlist, mandate/intent matching (against prompt-injection hijacks), AML screening, and anti-replay. KYA is the sharpest wedge; what's genuinely hard to copy isn't the number of gates but turning them into a protocol-neutral, any-agent-callable pre-settlement firewall we attack in public. We enter through identity — but we don't stop at it.
Not a slide deck — real numbers, and we broke it ourselves
78
6
3
Our KYA was once theater. In Attack Lab #3 we found that anyone who knew someone else's agentId could pay with it, zero credentials — “attributed” was a comforting lie. We published the result in the open, then added identity proof and one-click revocation. Whether a standard is worth trusting isn't about how complete it claims to be — it's about whether its authors publish the result when they break it themselves.
Want an ID card for your agent?
If you're building agents that spend autonomously, we're looking for a few early design partners to run KYA + the pre-settlement firewall together. Leave your contact and let's talk.