Know Your Agent: Payment Security When the Buyer Isn’t Human

The next wave of legitimate online shoppers won’t always look human. Some won’t type, move a mouse or even visit a traditional checkout page. They’ll be AI agents acting on behalf of customers.

For payment providers, that creates an entirely new security challenge. The automated behavior we’ve spent years learning to detect and block could soon represent genuine customers and valuable transactions. Get it wrong, and merchants could lose legitimate sales. Relax the rules too far, and fraudsters have a new opportunity to slip through.

The answer isn’t simply better bot detection. It’s knowing which agents to trust, what they’re authorized to do and who is accountable when something goes wrong.

Welcome to the emerging world of Know Your Agent (KYA).

In this article, we’ll examine the agentic identity problem, including what it means for your merchants, what the industry is doing to solve it and how to start preparing your services for an agentic environment.

If you’ve ever been asked to solve a CAPTCHA or “prove you’re human” during checkout, you’ve interacted with one of the most common security actions on the internet: filtering out bot traffic. Those filters exist because bot traffic runs rampant online. In fact, bots now make up over 50% of all web traffic.

In payments and ecommerce, every layer of fraud screening is designed specifically to detect and stop suspicious activity. But what happens when automation alone no longer signals hostile action? In the not-too-distant future, many legitimate customers will be bots, and that’s a problem that the payments industry (and your merchants) will need a solution to in a world of agentic commerce.

Until now, bot filtering has been mostly black and white: if a purchase exhibited any behavior that indicated it was generated by non-human traffic, it was flagged and likely blocked. Now, fraud screeners will need to operate in the gray, where good bot traffic, like artificial intelligence (AI) shopping agents, coexists with bad bot traffic. How will we tell good from bad? The behavioral signals that current fraud models rely on, including typing cadence, dwell time, mouse movement and navigation path, do not exist for agentic traffic. Step-up authentication like CAPTCHA, SMS one-time passwords and biometric prompts fails because it assumes a human is present. Malicious agents are also getting better at beating it. All this represents an enormous shift in the assumptions we currently use to screen for fraud.

Fraud screening faces two primary failure modes in an ecosystem of mixed bot traffic. The first is false-positive declines against good agentic shoppers. The second is improper approvals of malicious agents that seem like good ones. 

False Declines on Legitimate Agents

A high rate of false-positive declines is the most likely failure mode as agentic commerce rolls out. The reason is simple: trust in agentic payment systems is currently a problem. Merchants will keep their screening rules tight, and without additional identification, even the best AI-powered fraud screeners will struggle to tell good agents from bad.

So what happens when an agent runs into a false-positive decline? The Merchant Risk Council’s 2026 Global eCommerce Payments and Fraud Report found that false positives, known in the human shopping world as “customer insult,” are already widespread without any agentic traffic at all, with 65% of merchants putting their false-positive rate between 2% and 10% of orders and more reporting elevated rates this year than last. Agents change what a decline costs you. They take no offense and hold no grudge, so one bad experience will not sour them on a brand. But they have no loyalty either, and an agent that hits a decline will route the purchase to a competitor in the same instant. 

New Cover for Bad Actors

As agentic commerce grows in importance, merchants will face pressure to stop freezing out good agents. The easiest, most immediate solution would be loosening the rules that usually screen out non-human traffic in order to facilitate a higher success rate on agentic orders. That approach would reduce customer insult, but opening the gate wider lets more fraud through, too. That’s true even for rudimentary bots, but the real threat will come from advanced malicious agents that can disguise themselves as regular shopping agents. In that case, good agent traffic acts as a type of smokescreen that lets malicious ones blend in. 

Ultimately, part of the solution to keeping bad agents out while letting good ones in will come down to identity verification. But proving who an agent is will not tell us whether a particular purchase was authorized. Every agentic transaction will need to include proof that the buying agent:

• is a known, verifiable agent
• is acting on behalf of a verifiable person or business
• is authorized by that person or business to make this specific purchase within agreed limits

If you think back to the early years of ecommerce, the industry had to figure out ways to solve a similar identity problem with online payments that didn’t involve a physical card or face-to-face interaction. That led to systems like card verification value (CVV) codes, 3-D Secure and biometric verification that allow both merchants and consumers to trust doing business online. The agentic commerce ecosystem will go through the same technological evolution, albeit much faster.

KYA is the security framework taking shape for agent verification. Just like Know Your Customer (KYC) and Know Your Business (KYB) are designed to verify the identities of individuals and companies to curb illegal activities, KYA will give agents a verifiable and traceable identity and, equally importantly, tie that digital identity back to the person or business authorizing the agent. In my view, that is only the starting point. Payment security will also depend on proving what an agent was asked to do, what it was permitted to spend and how accountability can be established if something goes wrong.

How the Card Networks Are Tackling KYA and the Agent Identity Problem

In September 2026, Visa, Mastercard and Ant International, a Singapore-based fintech, announced a collaborative effort to advance KYA and establish industry-wide standards for verifying agent identity. 

Each of the three takes a different approach to establishing trust in agentic transactions: Visa Trusted Agent Protocol, Mastercard Verifiable Intent and Ant International Agentic Mobile Protocol. Mastercard co-developed Verifiable Intent with Google and designed it to work across other agentic protocols. Each represents a trust layer built to support agentic transactions, but each comes at the problem from a different angle. 

Visa’s Trusted Agent Protocol (TAP) verifies the agent at the store’s digital “front door,” passing a signed set of credentials that verify agent identity and authorization to act on behalf of the buyer. It tells the merchant the agent is trusted so it can pass through fraud screening.

Mastercard Verifiable Intent (VI) combines with Mastercard Agent Pay to create a secure chain of data that acts as a contract defining exactly what the agent is authorized to do and who authorized it. In a sense, it’s less about front-door access and more about creating a tamper-resistant record that can prove an agent’s identity and intent if something goes wrong. 

Agentic Mobile Protocol (AMP) is the trust layer developed by Ant International, the parent company of Alipay+. Unlike TAP and VI, it’s focused on digital wallet use instead of cards. It gives agents unique identities, verifies the activities they’re authorized to perform and assigns a proprietary trust rating that dynamically adjusts how much autonomy a shopping agent is given based on its behavior.

While the biggest players in payments are already hard at work solving the agent identity problem, it’s important to note that KYA is still in its infancy, and programs, protocols and naming are in a state of constant evolution. Your merchants are not waiting for it to settle. The same Merchant Risk Council survey found that 19% of merchants already have solutions in place to accept agentic AI payments, with another 63% planning or implementing them. Only 15% say they have no plans at all. 

The first wave of ecommerce taught us that being early is critical for capturing market share and taking advantage of new opportunities. So, with that in mind, the following are some action steps you can take to help make sure your merchants are ahead of the curve when agentic commerce starts to become a priority.

• Treat shopping agents as a distinct identity class: The binary paradigm of human or bot is over. You must now start planning to treat agentic shoppers as a unique monitoring class with its own thresholds and escalation rules. Consider tracking trusted agent attempts and false declines separately, so you can spot problems early.
• Audit your bot mitigation and velocity rules: Start digging into your rules-based screening systems to identify the triggers and thresholds that would cause a legitimate agent to be declined today. This understanding will be the foundation of the more nuanced bot screening your merchants will need. 
• Decide your position on delegated authority: You don’t have to accept the risks of external agentic shopping, as giants like Amazon have shown. You decide what level of delegation authority you allow your merchants to give external agents, including how a mandate is captured, scoped, monitored and revoked. That could mean spending limits, approved purchases and a straightforward way to withdraw permission. It’s a balance between enabling merchants and managing risk, so start defining the controls you need. 
• Revisit dispute handling: One of the biggest unanswered questions about agentic commerce is what disputes will look like. Liability for agent errors is still unsettled, and chargebacks could spike as end users work out how to control their agents. That means thinking now about what evidence merchants will need: the customer’s instructions, proof of authorization and a record of what the agent actually purchased. Clear dispute processes will matter as much as prevention.
• Educate your merchants: Trust will be the biggest source of friction in helping your merchants adopt agentic commerce. Ultimately, the more education you can start providing today, the more likely your merchants are to enter agentic commerce early and confidently.  

Agentic commerce won’t arrive overnight, but the merchants who are ready for it will capture the largest early gains. The goal is straightforward, but still challenging: let good agents through the gates without giving bad ones cover to sneak in alongside. KYA will give the industry the framework to make that happen. Standardization will take time, but the groundwork is something you can start laying today, so you’re ready to win early-adopting merchants tomorrow. 

Building that readiness starts with fraud defense that is native to your payments stack rather than bolted onto it. Today’s tools will not solve agent identity on their own, but strong controls and good risk data provide a solid starting point. At NMI, our partners get that foundation today:

• AI-powered fraud prevention through Kount 360, which uses machine learning to detect and block fraud in real time across billions of transactions
• Advanced security features including network tokenization, point-to-point encryption (P2PE) and 3-D Secure (3DS) to help protect digital transactions
• Native card testing detection built directly into your gateway
• Full Payment Card Industry (PCI) Level 1 compliance and a dedicated security team

To talk through what agent-ready fraud defense looks like for your portfolio, reach out to a member of our team today.