Why Agentic Commerce Demands Perpetual Risk Infrastructure

For most industries, agentic commerce is an interesting trend to watch. But for the payments industry, it's an existential pressure on some of the most foundational processes in the acquiring stack: onboarding and underwriting.

Legacy onboarding and underwriting workflows weren't built for this, and the gap between what those workflows can handle and what agentic commerce demands is about to become very visible. Fortunately, platforms like Worth deliver ongoing support and the perpetual KYC and KYB monitoring that financial institutions need to compete in an increasingly agentic world. 

The Rise of Agentic Commerce in B2B and B2C Payments

Imagine a customer opens an AI assistant and describes what they need. The assistant researches options, compares prices, selects a vendor, and completes the purchase without a human having to click a single button. Agentic commerce is already emerging across consumer and business contexts.

In April 2025, Mastercard and Visa both launched personal payment infrastructure specifically designed for AI agents. Mastercard Agent Pay and Visa Intelligent Commerce both use tokenization, authentication, and spending controls designed for transactions initiated without a human in the loop. Fast forward to 2026, and both companies have expanded their agentic commerce infrastructure to the business layer. 

Visa’s Intelligent Commerce Connect, launched in April 2026, is a protocol-agnostic on-ramp that lets agent builders, merchants, and payment enablers connect to agentic commerce through a single integration, supporting both Visa and non-Visa cards across major agent protocols. Mastercard’s Agent Pay for Machines, announced in June, focuses on infrastructure for machine-to-machine payments at scale, where AI agents transact with each other continuously at high frequency and low value, settled programmatically across cards, accounts, and stablecoins.

How Agentic Commerce Disrupts the Acquiring Stack

Traditional e-commerce has a human at the center of every transaction. A business owner browses, selects, enters payment information, and confirms. The entire acquiring stack, from application to approval to fraud monitoring, was designed around that assumption.

Agentic commerce removes the human from most of that loop. An AI agent acting on behalf of a business or consumer can research options, compare vendors, and complete a purchase without a person ever clicking a button. McKinsey projects that the global agentic commerce opportunity will reach $3 to $5 trillion by 2030, and the behavioral shifts are already measurable. Traffic to retail sites from GenAI sources increased 4,700% year over year as of mid-2025, per Boston Consulting Group (BCG) and Adobe.

The challenges this creates for acquiring are specific. Agents don't window-shop the way people do, hemming and hawing as consumers would. They execute their assigned tasks, which means transaction velocity can spike to levels that look indistinguishable from fraud under legacy pattern-matching models (think: rapid succession, identical amounts, off-hours activity). 

Merchants purpose-built to serve AI agents don't fit the standard profile that underwriting frameworks were built around: a business owner with a bank account. And when something goes wrong — maybe it’s an erroneous bulk purchase, a dispute rate spike, or a fraudulent transaction chain — the accountability runs through an AI agent rather than a person who signed your merchant agreement.

Each of these changes puts direct pressure on systems that weren't built to handle them.

Why Onboarding Can’t Keep Up

Current merchant onboarding workflows were designed around the assumption that a human is on the other end. Which, to be fair, is a historically valid assumption. Especially for those of us old enough to remember when you “had to” go to a bank in person to open an account (while online banking has technically existed since the mid-1990s, it wasn’t widely popularized until after the 2008 financial crisis). 

It’s a familiar story. A business owner fills out an application. A human reviewer or human-supervised system checks their identity, verifies their business, reviews their processing history, and makes a risk decision. This process typically takes days and involves multiple handoffs.

Agentic commerce disrupts this at two levels.

First, new merchant types are emerging specifically to serve AI agents — platforms, APIs, and micro-merchants that transact at machine speed and don't map cleanly to traditional risk criteria. The questions underwriters know how to ask (processing history, years in business, physical location) lose predictive value for merchants whose customers are autonomous agents executing at scale.

Second, the speed expectation is categorically different. Human merchants tolerate slow onboarding because they have to. They're invested in the outcome, they'll follow up, they'll wait for an email. An AI agent won't. If a vendor can't be verified and onboarded within the transaction decision window, the agent routes to the next available counterparty. There's no frustration, no follow-up email, no second attempt. Friction doesn't slow an agent down, but it does eliminate you from consideration.

Legacy onboarding workflows were never designed for this. A process built around document collection, manual review queues, and multi-day decisioning cycles assumes a counterparty that will wait. Agents don't wait around, and unlike human merchants, sometimes even unsolicitedly so, they won't tell you why they didn’t choose you.

Agentic Commerce Breaks Traditional Underwriting Models

Traditional underwriting has mostly been a point-in-time exercise. A merchant gets reviewed at onboarding, is ultimately approved or denied, and then operates with limited ongoing scrutiny unless something triggers a flag. That model has always had gaps. For example, the risk profile of a merchant at month one isn't the same as at month six. (That’s why Worth’s platform goes beyond onboarding and underwriting to perpetually monitor for ongoing risks.) 

Now, agentic commerce is positioned to magnify the gaps seen in old-school underwriting workflows.

When AI agents drive transaction volume, merchant risk profiles can shift quickly. A merchant that looked low-risk at onboarding might see dispute rates spike three weeks later, not because of fraud, but because an AI agent is purchasing at scale and the return rate is higher than anticipated. Batch-based monitoring and weekly exception reports can't catch this in time, and reactive chargeback programs aren't architected for the velocity that agentic commerce introduces.

The signals traditional underwriting relies on also lose predictive power. Credit history, years in business, and prior processing volume don't tell you much about a merchant purpose-built to serve AI buyers. New evaluation frameworks are needed. Frameworks that can dynamically assess risk based on real-time, perpetually monitored behavioral patterns, rather than static application data, such as Worth’s Continuous Risk Monitoring.

You KYB and KYC, but do you KYA?

Velocity anomalies are the visible symptom. The underlying issue is that the models watching for them weren't trained to detect agentic behavior. AI agents generate different fingerprints, and distinguishing legitimate agent activity from agent-driven fraud requires models trained on agentic transaction data, rather than patterns retrofitted from human behavior. 

An estimated 78% of financial institutions expect fraud to increase from AI shopping agents, and the manual review infrastructure that catches traditional fraud can’t scale to keep pace with the projected volume of agentic commerce.

KYC and KYB frameworks were built to answer the question: who is this person or business, and are they who they say they are? When the transacting party is an AI agent acting on behalf of a principal who may be several layers removed, that question becomes: who is ultimately responsible, and can you prove it? We’re entering the world of Know Your Agent (KYA).

Regulatory expectations have already begun to shift in this direction. The EU's Sixth Anti-Money Laundering Directive (6AMLD) and the incoming Payment Services Directive 3 (PSD3) are embedding KYC obligations directly into payment flows. The International Monetary Fund (IMF) recently published a policy note on how agentic AI reshapes payments, underscoring that this isn't a fringe concern.

Perpetual Fraud Detection and Compliance in an Agentic World

The volume of truly autonomous AI-initiated transactions is a small fraction of total commerce today; however, the trajectory is clear, and the technology enabling it is advancing rapidly. Companies that recognize this early and invest in risk infrastructure capable of handling machine-speed commerce will define the next generation of acquiring.

The question isn't whether agentic commerce will stress your infrastructure. It's whether your infrastructure will be ready when it does.

If you’re interested in learning more about how the Worth platform helps financial institutions adapt to evolving regulatory pressures, reach out to schedule a demo with our team.