Know Your Agent vs KYB in agentic commerce

Know Your Agent asks whether the software acting is authorised to act. KYB asks whether the business being transacted with is real and safe to transact with. They are frequently discussed as rivals, which is a category error.

The clean way to hold it: KYA verifies the messenger is authorised to carry the message. KYB verifies the sender and the recipient are legitimate.

Four questions define the layer. Authorisation: is there a human or organisational principal behind the agent, and can that authorisation be cryptographically verified? Scope: which actions fall inside the granted permissions? Integrity: has the agent's code or behaviour been tampered with since authorisation? Accountability: who is responsible if it causes harm?

These arise because commerce has always assumed a human or a known organisation on each side. An autonomous system breaks that assumption, and it matters most where agents are consumer-facing or payment-initiating — shopping agents, agents moving money, agents calling APIs for a user, and agents instructing other agents.

This is the distinction most often lost. When an AI processes a business's application, KYB requirements do not disappear — only the method of fulfilling them changes.

Four shapes recur: agent-driven onboarding, where an AI collects documents and queries registries; agent-to-agent commerce, where both questions apply at once; AI-assisted due diligence, where the agent's conclusion is only as good as the data it retrieved; and continuous monitoring for ownership changes, sanctions exposure and licence suspensions.

Side by side, the two layers differ in every dimension that matters:

This is the practical consequence, and the reason the frameworks are complementary rather than substitutable:

A KYA-verified agent reading poor business data still produces an unreliable answer. Agent authorisation and data quality are independent dimensions, and no amount of the first substitutes for the second.

Treating KYA as a replacement for KYB in agent-driven onboarding. KYA establishes that the agent is authorised and uncompromised; it says nothing about whether the business data the agent retrieved was accurate.

Treating traditional KYB as sufficient for agentic commerce. Agents can submit falsified documents, claim authorisations they do not hold, or operate outside sanctioned scope — none of which a registry check addresses.

Treating KYA as a consumer-only concern. Procurement agents submitting orders, contract agents signing agreements, and accounts-payable agents initiating transfers are all B2B, and all raise it.

At the agent layer: verify that agents reaching your platform are authorised by known principals, establish and enforce permitted actions, keep audit logs, and monitor for out-of-scope behaviour or credential misuse.

At the business identity layer: maintain fresh identity data, resolve entities so the identity is unified and correct, traverse ownership to reach beneficial owners and related-party risk, screen against sanctions and watchlists, and monitor for state changes that alter the risk.

Neither framework is mature. KYA standards are still emerging, and KYB infrastructure that answers at agent speed is still being built.