What is Know Your Agent (KYA)? AI fraud signals explained

A shopper asks an AI assistant to compare two laptops, watch for a price drop and build a cart once the price falls below $1,500. That kind of limited delegation is closer to how many people are using AI for ecommerce shopping today: Handing off parts of the buying process without necessarily handing over the purchase itself.

But that’s already changing. A 2026 Mastercard report predicts that by 2030, one in 10 online shoppers may use AI agents to both shop and pay. As agents take on more of the buying process, ecommerce merchants need to understand who’s acting on their customers’ behalf and what they’re authorized to do.

Know Your Agent (KYA) can help by giving retailers more information about the agent behind those interactions and the authority it has. For fraud teams, that adds useful context. But once an order is on the table, there’s still a harder question to answer: Is the transaction safe to approve?

How are AI assistants affecting ecommerce today?

Shoppers aren’t necessarily handing over their wallets and letting an agent complete every purchase. But they are regularly using AI to research products, compare options, monitor prices and build carts. 

New Signifyd data reinforces that pattern: Ecommerce sales originating from AI-assisted search jumped 815% year-over-year during the first half of 2026. That growth shows how much influence AI can have on a purchase even when the shopper is still the one completing it.

Visa’s own 2026 consumer research shows a similar gap. While 72% of U.S. consumers surveyed have used an AI assistant, only 23% said they trust GenAI to handle payment transactions on their behalf. And that willingness to delegate may not look the same for every purchase. Giving an agent standing permission to reorder a $20 case of wet cat food every three weeks is a very different proposition from letting it buy a $1,000 watch.

At the same time, agentic commerce isn’t waiting for every trust, identity and liability question to be settled first. Agents can increasingly work through existing ecommerce infrastructure, including conventional checkout forms and payment systems.

For ecommerce brands like yours, that means the trust question changes as agents take on more responsibility. As agent involvement moves from product discovery into account access, cart changes and checkout, you need stronger evidence around identity, consent and authorization.

You can think about agent involvement as a spectrum of delegated authority:

You don’t need the same level of proof when an agent recommends a laptop as you do when it accesses an account, chooses the model and submits payment. And when more shoppers do start having AI agents shop and transact on their behalf, the need for stronger identity, consent and authorization signals only grows.

What is Know Your Agent (KYA) in ecommerce?

In ecommerce, KYA is an emerging approach to verifying which AI agent is acting, who it represents and what it has been authorized to do.

Picture it as an identity and authorization layer that sits ahead of the buying decision.

For example, if an AI agent accesses a shopper’s account to build a cart, KYA could help you verify that it’s a known agent acting for that shopper and that it has permission to add items. But, that permission doesn’t necessarily extend to changing a shipping address or completing the purchase.

The easiest way to understand where KYA fits is to compare it with the identity frameworks you already know.

KYA vs. KYC vs. KYB at a glance

KYA, KYC (Know Your Customer) and KYB (Know Your Business) all apply identity verification, just to different entities.

What information can Know Your Agent give merchants?

Because KYA is still developing, the exact signal set will vary by implementation. But directionally, KYA could provide a few types of information:

Depending on the implementation, KYA signals could also show whether an agent’s credentials are valid, what limits apply to its authority and whether that authority has expired or been revoked. Or they may include more specific details, like spending caps, expiration times or the exact action the shopper approved.

Keep in mind: There isn’t one settled KYA standard or protocol today. Card networks, payment providers, agent platforms, security providers, commerce platforms and other technology companies are developing different ways to establish and carry those standards.

For example:

• Visa’s Trusted Agent Protocol gives participating agents a way to identify themselves to merchants and other commerce systems
• Mastercard’s Verifiable Intent creates a tamper-resistant record of what a shopper authorized an agent to do
• Google’s Agent Payments Protocol uses signed checkout and payment mandates to provide cryptographic proof of that authorization across agent-led commerce flows
• Personal Agent Protocol (Poppy) — an emerging open standard developed by Sierra, Meta, Stripe and other industry partners — aims to help businesses identify personal AI agents, authenticate their access and gain visibility into their interactions across websites, APIs and company-operated AI agents.

Those approaches differ, and you shouldn’t assume every signal will arrive neatly packaged at checkout.

Some identity and authorization checks may happen elsewhere in the ecosystem — between the agent platform, wallet, issuer or card network. You could receive some signals, different signals or potentially no explicit indication that an agent was involved.

Think about the dispute between Amazon and Meta over Muse: Amazon blocked the shopping agent from Amazon.com. Among its objections, Amazon said Muse didn’t identify itself while browsing the site.

That kind of visibility gap is exactly where KYA can help. But even when KYA signals are available, they still don’t tell you on their own if the resulting transaction is safe.

Does a verified AI agent mean a transaction is safe?

No.

Say a fraudster takes over an established customer account and then uses a legitimate shopping agent to place an order.

The agent could have valid credentials. It could be accurately identified. It could even be operating within the authority associated with that account. But, the problem happened before the agent ever got involved.

The same issue can show up elsewhere in the transaction. A legitimate agent tied to a real account could be given a stolen payment method. Or it could place an order that technically falls within its mandate but looks completely out of character for the shopper.

Consider two purchases submitted through the same verified agent:

The agent signal is the same. The surrounding transaction isn’t. 

KYA can inform a fraud decision, but it can’t make that decision on its own. Knowing that an agent is legitimate and authorized can increase confidence, but you still need to evaluate the shopper, account, payment method, order and surrounding behavior before deciding whether to approve the purchase.

How should merchants use KYA in fraud decisioning?

Treat KYA as another source of context, not a standalone approval signal.

Because AI agents can move through ecommerce sites at machine speed, create thin browsing sessions or operate from cloud infrastructure rather than the shopper’s usual device, behaviors that traditionally look suspicious may not be risky. To separate those legitimate AI agents from malicious automation, fraud teams need better context around the broader transaction.

At the same time, agentic commerce fraud can make questionable activity harder to spot because malicious agent-led behavior may look clean rather than noisy or obviously automated.

And on top of that, checkout infrastructure is being built to let agents complete purchases through your existing checkout, potentially without you knowing an agent was involved. So, you can’t wait for agent identity signals to arrive before assessing an order.

With that in mind, look at the evidence you have across the purchase:

• Shopper and account signals: Purchase history, known addresses, established payment methods and recent account changes.
• Agent signals: Agent identity, provider, credentials, authorization and mandate.
• Transaction signals: Order value, products, payment method, shipping destination, velocity and connections to known fraud.

Consider those signals together. A verified agent may increase confidence, while a password reset five minutes earlier, a newly added payment method or an unfamiliar shipping address could point in the other direction.

In other words, verifying the agent gives you another piece of the transaction story. You still need enough context to decide whether the story makes sense.

How does Signifyd fit into Know Your Agent and agentic commerce?

Signifyd fits at the decision, not the credential. While KYA can establish which agent is acting and what it’s authorized to do, Signifyd helps you assess the fraud risk behind the activity.

And that protection extends across the buying journey. For example:

• Session Intelligence identifies visitors by device and behavior from the first page load, helping you tell legitimate AI agents from bots and spoofed traffic before an account or order exists, even when an agent doesn’t announce itself.
• Account Integrity evaluates account creation, login and modification activity to detect suspicious behavior. This includes account takeover attempts (ATOs) that could compromise an otherwise legitimate agent-led purchase.
• Purchase Optimization uses AI-powered decisioning and intelligence from the Signifyd Commerce Network to evaluate identity and intent. This helps separate real purchases from fraudulent activity, including agentic orders that traditional fraud tools may mistake for bots.
• Guaranteed Fraud Protection evaluates orders at check out in real time, and backs eligible approved orders with a financial guarantee against chargebacks.

Does your fraud protection stack up against the challenges of agentic commerce? Request a demo to see how Signifyd’s solutions can empower your commerce decisions.