Every regulated financial system starts with a question about identity: who is this customer, and can we prove it? Know-Your-Customer grew up around people — documents, addresses, screening lists and a named account holder. That framework has served banks well for decades. Now a different kind of actor is transacting: software that acts on its own, on behalf of someone else, at volumes no human team could match. Regulators and standard-setting bodies are asking a fair question — can an identity framework built for humans stretch to machines?

What actually changed

An agent does not fit the old model. It owns nothing. It is not a customer in its own right. It can be revoked, replaced or retired. Meanwhile a person or a company still stands legally behind every operation. The new part is not that software moves money — that has happened for years — but that the software now decides.

Why KYC alone is not enough

KYC answers who the customer is. It does not answer whether this specific piece of software was permitted to make this specific payment, at this moment, within these limits. The identity of the principal and the authority of the agent are two separate questions. Treating them as one leaves a gap exactly where autonomy grows.

What a Know-Your-Agent framework needs

A workable framework tends to include:

  • an identified principal standing behind each operation;
  • a revocable agent identity that is never anonymous;
  • policy — limits, approved counterparties and approval thresholds — applied below the model;
  • partner rails for accounts, FX, payments and cards behind one interface;
  • a ledger with evidence, tracing each operation to its principal, agent and policy version.

Where several institutions are involved, a shared reliance agreement can help. FATF Recommendation 17 allows a regulated institution to rely on client due diligence performed by another, though it does not transfer legal responsibility — and cross-border reliance is not permitted everywhere.

In short

  • KYC identifies customers; it does not authorise software.
  • Agents are revocable, attributable and never anonymous.
  • Principal identification and agent authority are separate questions.
  • Reliance between institutions is possible but limited by local law.

This is general information, not legal, compliance or financial advice.

See how the framework is structured in the protocol.