Know Your Agent: The Bank Check AI Agents Need Next

Banks spent decades building KYC, know your customer. Now they need a second check, and it has a name. Know Your Agent, or KYA, is the idea that when software turns up at a bank to move money, the bank should be able to ask three questions before it says yes. Who built you, who sent you, and what are you allowed to do.
The phrase has floated around fraud vendors for a while. This week it got a regulator behind it. On 19 November, Chia Der Jiun, managing director of the Monetary Authority of Singapore, told the Singapore FinTech Festival that MAS will work on a common Know-Your-Agent protocol. Per Asian Banking & Finance, customers and banks would swap details on an agent's identity, who it represents, its authority, the mandate it works under, and its provenance. A day earlier, MAS and the Future of Finance Institute announced an Agentic Money Sandbox with DBS, OCBC, Standard Chartered and UOB as launch partners. It will test agent-initiated payments and treasury work "within defined authority, limits and safeguards."
Why Know Your Agent matters now
The scary version has its own nickname: the agentic bank run. In September, Apollo's chief economist Torsten Slok wrote a short note called Is an Agentic Bank Run Coming? His point was plain. US checking accounts pay a 0.1% national average while other accounts pay 3.3% to 5.0%. If every household had an agent sweeping cash toward the better rate, banks would lose the cheap deposits they lend from.
Then on 16 November, Fed Governor Michael Barr went further in a speech in Milan. He said major banks now report AI agents trying to act in retail accounts "without the bank being able to verify authenticity, consent, and alignment with customer instructions." He also warned that agents could "autonomously move funds at high speed, potentially accelerating runs that unfold in minutes."
Minutes. Old-style runs needed a queue outside the branch. The 2023 ones needed a group chat and a panic. An agent needs a rate gap and a cron job.
A quick example
Say I give my assistant one job: keep my savings in the best-paying account. Today the lazy way is to hand it my net banking password. The bank then sees "Mani" logging in, when it is really a bot acting for Mani, and nobody has a record of what I actually asked for.
With KYA, the agent shows up with a signed card instead. Something like: built by vendor X, acting for Mani G, may move up to ₹50,000 a week between these two accounts, cannot add new payees, mandate ends in March. The bank checks the card, enforces the cap, and can slow things down if ten thousand agents try the same move at 3 a.m.
That is the whole trick. It is OAuth scopes with a compliance officer bolted on. Not glamorous, but most of finance is not.
My take
I am for it, and I think it is late. Password sharing with bots is the real risk today, not some sci-fi swarm. Barr's own fix list is the right one: strict authentication, explicit consent for specific actions, and hard technical limits on what agents can do. Chia put it in four words: "Autonomy has to be bounded."
My worry is the gap between the protocol and the bill. A signed mandate tells the bank what the agent may do. It does not settle who pays when the agent does something dumb inside its mandate. Expect that fight to take longer than the spec.
If you build agents that touch money, start now. Log every mandate, keep read access apart from action access, and make the agent say who it works for. Regulators are going to ask.
So here is my question. Would you let an agent move your money if the bank could see its mandate, or is a human tap on "confirm" still your line?