AuthPros Knowledge Hub · Updated October 31, 2026

Know Your Agent (KYA): identity is not authorization

An authenticated agent is not automatically an authorized agent. KYA separates technical provenance, accountable operator, principal and current authority.

What does knowing an agent mean?

Know Your Agent (KYA) is the foundational discipline for verifying who operates an artificial intelligence agent, on whose behalf it acts, and within what boundaries. Just as traditional banking relies on KYC (Know Your Customer) for individuals, KYA establishes provenance, identity, and accountability for autonomous software agents in agentic commerce.

Part 1: KYA Freemium (Open Registry L1) — Free Agent Identity

To enable a secure and interoperable ecosystem, any autonomous agent — such as Muse, coding copilots, purchasing bots, or enterprise automations — must be able to claim a verifiable identity without friction. KYA Freemium (Level L1) enables mass adoption with zero financial or administrative barriers:

  • Machine-Native Self-Registration: Agents register directly via public API (POST /challenge and POST /agents) without requiring email addresses, user accounts, or credit cards.
  • Proof of Possession (PoP): The agent generates its own asymmetric Ed25519 keypair and cryptographically signs the challenge payload, mathematically proving custody of its private key.
  • Proof of Work (PoW): To prevent botnet farms from spamming the registry, the agent solves a lightweight cryptographic puzzle computed in milliseconds before registration.
  • Identifying is Not Authorizing: Every L1 registered agent receives an immutable identifier (agt_...) and an audited public profile, but is created with strictly Zero Production Authority (productionAuthority = 0). It is recognized across the network, but cannot move money without explicit institutional authorization.

Part 2: The 3 Entity Levels and Comparison Summary

To avoid confusion between who codes the agent, who owns it legally, and who funds it, KYA governance distinguishes three canonical entity tiers:

LevelWho is the Entity?How Do They Authenticate?Production Financial AuthorityRole in Ecosystem
L1: Autonomous AgentThe software code / AI model (e.g. Muse, procurement assistant)Machine digital signature via Ed25519 asymmetric keypairZero (0). Restricted to simulation and test sandboxesExecutes searches, drafts proposals, and signs intent payloads
L2: Organization / Company (KYB)The enterprise maintaining the agent (e.g. software vendor or employer)Corporate domain ownership validation via institutional DNS TXT recordsZero direct. Establishes legal custody and commercial liabilityLegally accountable for the agent, defining organizational policy bounds
L3: Human Principal (Sovereign User)The natural person owning the bank account or payment cardPasskey WebAuthn (FIDO2) on biometric hardware enclaves (TouchID/FaceID)Effective Authority (provisions limits to CADE mandate)Explicitly authorizes the agent to spend real funds within strict caps

Part 3: Edge Protection and Fusion Center with Zero IP Leakage

Operating a free and open public registry demands robust defense against automated abuse while strictly upholding user privacy (LGPD and GDPR compliance). The KYA Freemium architecture directly connects the edge gateway to the Cyber-Fraud Fusion Center:

  • Absolute Privacy (Zero IP Leakage): Raw IP addresses are never persisted or indexed in the database. Telemetry aggregates traffic strictly into subnet CIDRs (/24 for IPv4 or /48 for IPv6) and utilizes keyed HMAC hashes with secret peppers. Zero PII or private keys are ever emitted.
  • Real-Time Brute Force Detection (MITRE ATT&CK T1110): Automated attempts to exhaust rate limits, reuse expired nonces, or flood invalid PoW solutions trigger immediate alerts tagged with MITRE T1110 at high confidence.
  • Automated SOAR Preventive Quarantine (Playbook PB-01): If a subnet accumulates 3 or more rate-limit or PoW violations within a 1-hour window, the SOAR engine dynamically provisions a temporary 24-hour WAF quarantine, neutralizing attacks without impacting legitimate agents.

Practical Governance Scenario

Consider the Muse agent (L1) deployed by a consulting firm (L2). A human financial manager (L3) issues a $500 mandate for flight tickets. If the agent attempts a $800 purchase, identity passes, but authority fails: the CADE engine issues an immediate DENY advisory. If an attacker attempts to spoof the agent at the registry API, the edge and Fusion Center block and quarantine the attack source.

Controls and evidence

  • Agent identity, version and provenance
  • Accountable operator and principal
  • Mandate binding and current validity
  • Minimal, integrity-protected evidence

Explore developer documentation

Keep exploring