There is a question that nobody had to ask ten years ago, barely anyone was asking five years ago, and that the entire financial industry is now scrambling to answer at speed.
Not "who are you?" That one we more or less figured out. Passports, passwords, biometrics, two-factor authentication, the works. Decades of infrastructure built around verifying that you are who you say you are.
The new question is this: who is your agent, and can we trust it?
Because here is what is happening right now. AI agents are shopping on your behalf. Not browsing and making suggestions for you to approve. Actually shopping. Comparing products, reading reviews, selecting items, plugging in payment details, completing transactions. The whole thing, start to finish, while you are making breakfast or sitting in a meeting or sleeping.
And the merchants, the payment networks, the fraud detection systems, and the banks on the other end of those transactions are staring down an entirely new kind of entity arriving at their checkout. Not a human. Not a traditional bot. Something in between, and something they have no established playbook for.
That gap, between "agent shows up to buy something" and "merchant confidently accepts the transaction," is where everything interesting, and everything risky, currently lives.
The Old World Was Simple Enough
In traditional e-commerce, the chain of trust was relatively short.
You, a human being with a verified identity, went to a merchant website. You entered your card details. The card network checked with your issuing bank. The bank confirmed the card was valid and the funds existed. The merchant processed the order. Done.
There were fraud problems, obviously. There still are. But the fundamental question of who was on the other end of the transaction had a fairly clear answer: a person, using their own card, on their own device, at a known IP address.
Now stretch that chain out. An AI agent, running on a platform, powered by one or more language models, possibly using third-party tools for browsing and discovery and product comparison, arrives at a merchant checkout. It wants to place an order using payment credentials that belong to a human being who is not present.
Suddenly the questions multiply fast. Is this agent legitimate? Which platform is it running on? Who is the human it represents? Did that human actually authorise this specific transaction? Has the payment credential been stolen and handed to a fraudulent agent? Is this even an agent, or is it a scraper bot pretending to be one?
The merchant does not know the answers to any of these questions. Neither does the fraud detection system. Neither does the card network. And in that uncertainty, bad things can happen to everyone involved.
Enter KYA: Know Your Agent
You have probably heard of KYC. Know Your Customer. It is the bedrock of financial compliance, the process by which banks and financial institutions verify the identity of the people they do business with. Name, address, date of birth, identity documents, the whole ritual. It exists because anonymous money movement is a gift to criminals, and regulators decided decades ago that financial institutions had to actually know who they were dealing with.
KYA is the same idea, applied to the new world.
Know Your Agent means building infrastructure that can answer, reliably and verifiably, the layered identity questions that now sit at the heart of every agentic transaction. Who built this agent? Who is it running on behalf of? What has it been authorised to do? And how do we know any of this is true rather than fabricated?
It sounds simple. It is genuinely hard.
The Layers of the Problem
Here is where it gets interesting, because agent identity is not a single question. It is a stack of questions, and you need good answers at every layer.
Layer one: the platform.
Who built and operates the platform this agent is running on? Is it a reputable company with proper security practices and accountability? Has it been vetted? This is the foundation. If the platform itself is untrustworthy, nothing built on top of it is worth much.
Layer two: the agent.
Within a platform, different agents can have very different levels of capability and authorisation. The identity of the specific agent matters, not just the platform it lives on.
Layer three: the principal.
This is the word the industry is starting to use for the human or business that the agent is acting on behalf of. The principal is the actual customer, the person whose money is being spent, whose interests are supposed to be served. Knowing who the principal is, and being able to verify them, is crucial.
Layer four: the mandate.
What has the principal actually authorised this agent to do? Is it a broad standing instruction, like "handle my grocery shopping"? Or a specific one-time permission, like "buy these exact items from this merchant for under this amount"? The mandate is the scope of the agent's authority, and it matters enormously when something goes wrong.
All four of these layers need to be verifiable, communicable, and trusted by the parties on the receiving end of the transaction. That is the KYA challenge.
Why Merchants Are Genuinely Worried
If you talk to merchants about agentic commerce, and many people in the payments industry are doing exactly that right now, one concern comes up over and over again.
Disintermediation.
For as long as e-commerce has existed, merchants have worked hard to build direct relationships with their customers. They know who buys from them. They can market to them, retain them, delight them with good experiences, and learn from their behaviour. That direct relationship is enormously valuable.
An AI agent threatens to dissolve it.
If a consumer never visits the merchant's website, never sees the brand, never interacts with the checkout experience, just receives a parcel one day because their agent decided this was the best place to buy from, the merchant has lost something important. They have become, in the bluntest possible terms, a fulfilment operation for someone else's platform.
This is not a small concern. It is an existential one for many businesses.
And so merchants, quite reasonably, are demanding to know who the agent is. Not because they want to block AI agents from shopping with them. They very much want the sales. But because they want to maintain some connection to the end customer, some visibility into who is buying from them, and some protection against the scenario where they are simply serving anonymous traffic that could evaporate at any moment.
KYA infrastructure, by revealing who the principal is behind every agent transaction, is one of the key mechanisms for giving merchants that visibility back.
The Bot Blocker Problem
Here is something that most people outside the industry do not think about, but that is a very live issue right now.
Before an AI agent can even get to checkout, it has to get past the front door.
Most major e-commerce sites run bot management software. Systems designed to detect and block automated traffic that is scraping prices, hoarding inventory, launching fraud attacks, or doing any of the other unpleasant things that malicious bots do. These systems are good at their jobs. And they do not particularly care whether the automated traffic arriving at their door is a malicious scraper or a well-intentioned AI shopping agent trying to buy running shoes for someone.
From the bot blocker's perspective, it all looks the same: non-human traffic, suspicious behaviour, block it.
This means that a huge proportion of AI agents trying to shop on the legitimate web are currently hitting 403 errors. Access denied. Not welcome here. Go away.
Solving this requires a way for agents to identify themselves to these systems in a way that is both verifiable and trusted. Not just claiming to be a legitimate agent, because anyone can claim anything. But actually proving it, cryptographically, with credentials that have been issued by a trusted party and that carry verifiable information about the platform, the agent, and the principal behind it.
When an agent can walk up to a bot blocker and say, in effect, "I am an agent running on this verified platform, acting on behalf of this verified user who has a valid payment credential issued by this bank, here is the cryptographic proof of all of that," the bot blocker can make a real decision. Let it through. The commerce can happen.
Without that identity infrastructure, agents either get blocked, or bot blockers get loosened in ways that let actual bad actors through. Neither outcome is good.
What a Trusted Agent Payload Actually Looks Like
When a verified agent makes a request, the payload it carries, the information it sends along with every HTTP request, looks something like this in terms of what it communicates.
The platform that built and operates this agent. The specific agent instance. The identity of the principal, verified by KYC processes at the point of onboarding. The payment credential, tokenised so the raw card number is never exposed. The mandate, the scope of what the agent has been authorised to do. And a cryptographic signature tying it all together so none of it can be forged or tampered with.
The merchant, or the bot blocker in front of the merchant, receives this bundle and can make an informed decision. This is not an anonymous bot. This is a verified agent, acting for a verified person, with a verified payment method, operating within a defined authorisation. Let it through.
That bundle of verified information is, in a sense, the agent's passport. And just like a passport, its value depends entirely on the trustworthiness of the institution that issued it.
The Consent Question Nobody Is Ignoring
Running through all of this is a question that sounds philosophical but is actually very practical.
Did the consumer actually agree to this?
Not in the vague sense of having once installed an app and tapped through a terms and conditions screen. In the specific, auditable, legally meaningful sense of having authorised this specific type of transaction, with these parameters, through this agent, on this platform.
The protocols being built right now take this seriously. The approach that is emerging involves capturing user consent not as a one-time blanket permission but as a structured, scoped mandate. You authorise the agent to spend up to a certain amount, in a certain category, within a certain time window. The agent operates within that mandate. The consent is logged with a timestamp and tied to a biometric or passkey verification on a bound device.
When a dispute arises later, and disputes will arise, that logged mandate becomes the source of truth. The network can say, here is what you authorised, here is the timestamp, here is the verification. This was within scope. The transaction was valid.
This matters not just for fraud prevention but for consumer protection. In a world where agents are making purchase decisions without your direct involvement, the audit trail of what you actually consented to is your primary safeguard.
The Bigger Picture
KYA is not just a technical challenge. It is a foundational question about how trust gets established in a commerce ecosystem that has fundamentally changed shape.
For thirty years, trust in e-commerce was built around the human buyer. Their verified identity, their conscious decision at checkout, their card details entered by their own hands. All of the fraud rules, liability frameworks, consumer protections, and dispute processes were built around that model.
That model is being replaced. Not gradually, but rapidly. And the infrastructure of trust needs to be rebuilt around the new model before the old one fully disappears.
The good news is that people are working on it, hard, right now. Payment networks, fintech companies, AI platforms, consumer advocates, and standards bodies are all in the room together trying to answer the same questions. What does agent identity look like? How do you verify a principal? What constitutes valid consent? Who is liable when something goes wrong?
The answers are coming. The protocols are being drafted. The standards are being set.
But here is the thing worth sitting with.
The standards being set right now will define, for a long time, how much power AI agents actually have over your financial life, how accountable they are when they get things wrong, and whether you, the consumer, are genuinely protected or just conveniently assumed to have consented to whatever happened.
That is not a question to leave entirely to the engineers and the card networks, as talented and well-intentioned as many of them are.
It is a question for all of us.
Because your agent is coming. And the question of whether it truly knows you, and whether anyone can truly know it, is one of the most important design decisions being made in commerce right now.
This piece draws on discussions from the Loyal Agents Initiative, a collaboration between Stanford HAI's Digital Economy Lab, Stanford CodeX, and Consumer Reports Innovation Lab.
This Is Exactly the Problem Roving Is Building For
Everything you just read, the identity questions, the layered trust stack, the bot blockers, the mandate structures, it all points to one thing that does not yet properly exist.
A financial layer built specifically for AI agents.
Not a workaround. Not a tokenised card credential passed through a chain of intermediaries and hoped for the best. A proper banking infrastructure where every agent has its own verified financial identity, its own account, its own controlled spending card, and a complete audit trail of every single action it takes with your money.
That is what Roving is building.
Roving is the neo bank for the agentic economy. Give your AI agent a real bank account, smart virtual cards, spending rules it cannot override, global payment capability, and a verified KYA identity before it moves a single penny. Whether you are building agents, deploying them across a business, or just watching this space with growing interest, Roving is where the financial infrastructure for that future is being built right now.
Get early access at roving.money and be among the first to give your agents the financial identity they need to operate safely in the world.
If this piece made you think, drop a comment below. We want to hear from you.
Stay in the loop:
Follow us on LinkedIn for product updates, industry news and founder notes
Join our Discord community and be part of the conversation shaping the future of agentic finance
Share this with someone who needs to read it, the developer building agents, the fraud team rethinking their rules, or the product manager who has not yet thought about what KYA actually means for their stack