Know your agent: identity for software that acts for people
Know your agent is industry usage, not law. It combines the customer's proofed identity, the agent operator's verifiable identity and a link between them.
"Know your agent" borrows its name from know-your-customer, but it is industry usage, not a legal requirement. It describes checking three things before an agent may act: which agent this is, who operates it, and whom it acts for.
The customer side is established practice. Identity proofing at onboarding, described in NIST's Digital Identity Guidelines and in financial regulators' customer identification rules, establishes who the customer is. The Federal Reserve's SR 26-6 answers questions on using verifiable digital credentials under the US customer identification programme rule, and the EU's digital identity wallets will give customers a strong, portable way to prove identity and authenticate.
The agent side is newer. Building blocks include:
- Signed requests. With HTTP Message Signatures, an operator signs its agent's requests so a site can verify them. The IETF Web Bot Auth working group is standardising this for automated clients.
- Agent descriptions. Protocols such as Agent2Agent publish agent cards describing capabilities and authentication requirements.
- Scheme programmes. Card schemes register agents and issue them their own tokens.
The link. The decisive piece is binding the two: a mandate or token that names the agent and the principal together, issued after the customer authenticated strongly.
A know-your-agent decision should be logged as an outcome, allow, limit, challenge or refuse, naming the agent, its operator and its principal.
Sources: NIST SP 800-63-4: Digital Identity Guidelines, Federal Reserve SR 26-6: FAQs on verifiable digital credentials under the CIP Rule, European Commission: European Digital Identity Regulation, IETF Web Bot Auth (webbotauth) Working Group, RFC 9421: HTTP Message Signatures