
Know Your Agent (KYA) is the process banks use to check AI agents that act for customers. Say a customer tells an AI assistant to pay the electricity bill and move what’s left into savings. The assistant logs in and makes both payments without the customer opening their banking app.
Before allowing that, the bank needs to know the assistant is genuine, that the customer approved it, and what it’s allowed to do. Banks already ask those questions about people under Know Your Customer (KYC) rules. KYA applies them to software.
What Does Know Your Agent Mean in Banking?
Know Your Agent is the set of checks a bank runs on an AI agent: which customer approved it, which company built it, and what it’s allowed to do.
Banks already run similar checks on people. Under KYC rules, they check ID, screen sanctions lists, and rate each customer’s risk. Many banks track that relationship over time with client lifecycle management software.
It holds identity records, risk ratings, and review dates in one place. KYA adds a new entry to that record: the agents a customer has approved and the limits on each one.
Why Aren’t Current KYC Checks Enough for AI Agents?
Current KYC checks aren’t enough because they assume a person is behind every transaction, and a login can’t show whether software is acting instead. A password and a one-time code count as proof that you chose to pay.
Customers have used digital assistants like Siri and Alexa for simple tasks for years. Agents go further. They can log in, compare options, and send money with little human input. If a customer gives an agent their password, the agent passes every login check. The bank can’t tell that software is in control.
An agent with valid access can also be tricked. Agents read web pages as they work. Hidden text on a page could tell one to send money the customer never meant to send. In April 2026, the IMF warned that agent-made payments can blur who approved a transaction and who is liable when it goes wrong.
How Will Banks Verify an AI Agent?
Banks will verify an AI agent in five steps: verify the person, register the agent, link the two, set limits, and watch every action.
- Verify the person. The customer passes normal KYC checks, like an ID scan and a face match.
- Register the agent. The company that built the agent registers it with the bank or card network. The agent gets a digital credential, which works like a signed ID card that is hard to copy.
- Link the agent to the customer. The credential points to a verified person, not just an account number.
- Set limits. The customer decides what the agent may do. A bank might let it read balances but ask the customer to approve any large transfer.
- Watch every action. The bank logs each instruction and flags anything outside the limits. The customer can cancel the credential at any time.
This is the biggest change from KYC. Customer checks mostly happen at sign-up and on set review dates. Agent checks need to run on every transaction.
What Are Visa, Mastercard and Others Building?
Visa and Mastercard are building systems that register AI agents and give them credentials that banks and stores can check before a payment goes through. Most agents today help people shop, so card payments are where these checks are being built first.
- Mastercard Agent Pay (April 2025): Agents must be registered and verified before they can pay. Payments use tokens in place of real card numbers.
- Visa Trusted Agent Protocol (October 2025): Built with Cloudflare, it lets approved agents prove who they are to merchants with digital signatures.
- Shared KYA framework (September 2026): Ant International said it is working with Mastercard and Visa on common rules. The work runs through BuildFin.ai, a platform set up by Singapore’s central bank.
There is no single global standard yet, so banks will likely need to accept credentials from several systems. Regulators also haven’t set formal KYA rules.
How Should Businesses Prepare for KYA?
Businesses should prepare by working out which side of an agent transaction they’re on. A business may build agents, serve customers who use them, accept payments from them, or use agents for its own finances. Each role needs a different step.
If you build AI tools that handle money, expect banks to ask who made your agent and how it’s secured. They will also want to know what it’s allowed to do and how a customer can switch it off. Building in registration, clear permissions, and activity logs now costs less than adding them later.
If you run a fintech, lender, or payments business, your own customers will start sending agents your way. Record which agents each customer has approved alongside their KYC details. Then check agent activity on every transaction, not only at onboarding.
If you sell online, ask your payment provider which agent credentials it accepts. That way, real buyers using agents don’t get blocked as bots.
If your team uses AI agents for company finances, connect them through tools your bank has approved instead of sharing logins. Give each agent only the access its task needs.
Will Banks Let AI Agents Act for Customers?
Yes, but only agents they can trace to a verified person and hold to clear limits. The banks best placed to say yes will check agents on every action, not just once at sign-up.


