KYA— Know Your Agent

Know Your Agent

Trust, identity, and governance for autonomous agents.

Autonomous software should not operate anonymously. Every agent needs identity, authority, limits, oversight, and auditability — by design.

Explore the protocolRequest early access
  • Identity
  • Ownership
  • Authority
  • Oversight
  • Audit
01
Agent
02
Identity
03
Owner
04
Risk
05
Policy
06
Approval
07
Execution
08
Audit
IntentVerificationDecisionExecutionRecord

Why KYA

Agents can act. Someone must trust them.

Autonomous systems need more than capability — they need identity, ownership, authority, and oversight to be safely useful.

Identity

A stable, verifiable identity for autonomous software — tied to a prompt fingerprint and a vendor profile.

Ownership

Every agent links back to a responsible owner or principal. Actions remain attributable.

Authority

Agents operate only within defined mandates and policy limits — never with open authority.

Oversight

Critical actions remain reviewable, interruptible, and auditable end-to-end.

KYC → KYA

KYC verifies people. KYA verifies agents.

The same governance principles, extended to autonomous software. Owners remain accountable; agents operate under defined authority.

KYC

Verifies people
  • Human identity
  • Beneficial owner
  • AML profile
  • Account access

KYA

Verifies agents
  • Agent identity
  • Owner / principal linkage
  • System prompt hash
  • Data sources
  • Human oversight flag
  • Spend authority
  • Risk score
  • Velocity limits
  • Execution permissions

Profile

What defines an agent?

A KYA profile binds an agent to a vendor, a prompt fingerprint, a stated purpose, and a measurable risk posture. The result is a record auditors and owners can both read.

  • Model and vendor — what is the agent built on?
  • Prompt fingerprint — what behaviour is it locked to?
  • Data sources — what does it have access to?
  • Risk score — what is its measured exposure?
  • Spend cap — what is its authorised ceiling?
  • Oversight — is a human in the loop?
KYA profile
approved
agent_id
agt_01HEZ8K3W9XM2
model
gpt-4o
vendor
openai
purpose
Procurement automation
prompt_hash
sha256:9a4f…b21c
data_sources
public_web, internal_erp
spend_cap
5,000 USD / day
risk_score24 / 100low
human_oversightenabled

Lifecycle

KYA is a lifecycle, not a checkbox.

From draft to active to revoked, every transition is recorded and reviewable. Authority can be granted, frozen, or withdrawn at any point.

  1. 01
    Draft
    Agent is created locally; KYA profile not yet submitted.
  2. 02
    Pending review
    Submission accepted; compliance review in progress.
  3. 03
    Approved
    KYA decision recorded. Agent may begin authorised actions.
  4. 04
    Active
    Operating under policy limits; every action is audit-logged.
  5. 05
    Frozen / Revoked
    Authority withdrawn. Pending actions are blocked.

Safety rails

Autonomous systems require hard limits.

Caps, velocity, approval, kill-switch, audit — five guardrails that contain agent behaviour even when models behave unexpectedly.

Spend caps

Hard daily and per-action ceilings, enforced at execution time.

Velocity controls

Bounded action rates over rolling windows. Anomalies pause execution.

Human approval

Sensitive proposals require an explicit confirmation by the owner.

Kill switch

Owner-initiated freeze immediately stops new actions across rails.

Audit trail

Append-only records of proposal, approval, execution, and outcome.

Operators

Built for compliance reviewers.

A focused workspace for the humans who approve, reject, and audit agent submissions — with risk context already on the screen.

  • Triage queue ordered by risk and age
  • Side-by-side profile and prompt fingerprint
  • Approve or reject with one keystroke
  • Append-only review history
See the operator workflow
kya / operators / review
live
Search agents, owners, fingerprints…

Early access

Trust is the missing layer in agentic systems.

Build agents that can act — with identity, limits, approvals, and proof.

Request early accessRead the protocol