Searching for Trust in Agentic Commerce

When an AI agent buys the wrong product—or makes a purchase no one explicitly approved—the fallout isn’t just a customer service issue. It’s a liability problem the payments ecosystem isn’t fully prepared to handle.

In a PaymentsJournal Podcast, Jill Willard, CTO at IXOPAY, Rory Herriman, CTO and COO at Zip Co, and Christopher Miller, Lead Analyst of Emerging Payments at Javelin Strategy & Research, explored how liability may evolve as AI agents take on more responsibility in transactions.

A Multidimensional Problem

That question of liability quickly becomes a technical one: how do you even evaluate trust in an agent that isn’t human?

The first challenge is determining how to calculate a trust score for an AI agent that lacks a traditional human behavioral footprint. Any viable framework must extend beyond identity to include intent—and how that intent translates into behavior.

This isn’t just an engineering challenge; it’s also a cognitive one. Professionals in this space must rethink how they evaluate risk, developing new instincts that help them focus on the right signals.

“We have certain models, frameworks, and even language that professionals use to describe the vectors of risk,” Miller said. “As we think about the replacement of human actors with agentic actors, we lack instinct. The mere notion blocking bot traffic as a way of defending against fraudulent behavior stops being useful. It becomes anticommercial.”

Herriman added: “We have to approach it as a multidimensional problem. “It's not just ‘Is this actor good?’ Even if the actor is good, there are other dimensions on top of the binary switches, a complexity that never disappears.”

Assigning Liability in a New Context

In the world of AI, merchants are steadily losing control over the checkout experience. Decision-making has shifted upstream. A consumer can now instruct agent not only to buy “blue shoes,” but to purchase them from a specific merchant.

This shift brings a corresponding liability. If an agent buys light blue shoes instead of dark blue, who is at fault? Today, that burden often falls on the merchant.

“We've seen some card brands, such as Amex, say that they're going to accept liability for agentic transactions, which is an awesome development,” said Herriman. “But even if the liability fully shifts and the card brands take on more of that liability, there's still a cost to shipping the wrong goods out. That can be the hard cost of the shipping fees or the operational cost to get the goods out the door, but it can also be at the cost of customer relationships.”

Liability will become a central issue for issuers, providers, and merchants in the coming years. Existing frameworks address stolen cards or unauthorized use of payment credentials. But when a consumer is dissatisfied with what an agent selected on their behalf, responsibility becomes far less clear.

A similar cycle emerged in early e-commerce. Merchants drove growth by offering generous return policies and absorbing the associated risk. Over time, return rates skyrocketed, leaving businesses with inventory that couldn't be resold at full value.

Eventually, that broad assumption of liability narrowed. Companies began analyzing customer behavior and limiting privileges for high return users—an early example of risk-based personalization.

Responding to Complexity

Payments have never had a single, unified approach, and agentic commerce is no exception. Agents will operate across multiple protocols and may express preferences for how transactions are executed—for example, specifying which rewards card to use.

Orchestrators are working to simplify this complexity for merchants, who are primarily focused on selling products—not managing payments infrastructure. Meanwhile, agent developers are not necessarily optimizing for merchant interests.

“Merchants have to figure out how to participate in an ecosystem where they aren't necessarily the reason why the products have been developed