Page 18 - gs260702

Insights and Expertise

Why agentic The system is imperfect, but it follows a recognizable logic.

commerce is Agentic commerce introduces a fourth scenario that does
not fit neatly into any of those categories. The card was not
stolen. The merchant did not necessarily make a mistake.
creating a new The AI agent may have acted exactly as authorized. Yet
the customer looks at the transaction and says: "I didn't
want that."
category of dispute Perhaps the agent renewed a subscription automatically.

Perhaps it selected an alternative product, reordered an
item no longer needed or booked travel that technically
matched a schedule but not the customer's preferences.
From the system's perspective, the transaction may ap-
pear entirely legitimate. From the customer's perspective,
however, the outcome feels wrong. And when a transac-
tion feels wrong, many consumers instinctively turn to
the dispute process.

Traditional chargeback evidence such as authentication
checks, delivery confirmation and device fingerprints was
designed around human behavior at the point of transac-
tion. Agent-initiated commerce requires something funda-
mentally different: evidence of what the customer autho-
rized the agent to do, what limitations were in place, how
decisions were made and when consent was provided.

That evidential framework does not currently exist by de-
fault. Merchants, issuers and platforms will need to build
it deliberately.
By Monica Eaton
Chargebacks911 The problem no one is talking about
Much of the industry conversation has focused on the ob-
or fifteen years, I have watched the chargeback vious risk: consumers disputing AI-initiated purchases
landscape evolve in ways that much of the pay- they genuinely did not intend to make. That challenge is
ments industry failed to anticipate. Friendly real and it will grow. But there is another issue emerging
F fraud, first-party misuse, consumers disputing much earlier in the transaction lifecycle that deserves far
transactions they simply do not recognize. Each shift cre- more attention.
ated new pressure on merchants, new blind spots in fraud
systems and new questions around who should ultimately Legacy fraud systems are already blocking legitimate
bear the cost. agentic transactions before they are completed. Most
fraud tools were built to detect suspicious human behav-
What is happening with agentic commerce is categorically ior through unusual devices, abnormal purchasing pat-
different. AI agents are rapidly moving from recommend- terns or inconsistent session activity.
ing purchases to executing them. Visa has already com-
pleted secure agent-initiated transactions with partners, A legitimate AI agent and a malicious bot can appear re-
while Mastercard is expanding its own agent payment in- markably similar to systems that were never designed to
frastructure through Agent Pay. distinguish between the two. The challenge is no longer
simply detecting fraud, but differentiating between ma-
Platforms including Perplexity, Amazon and Walmart are licious automation and legitimate delegated purchasing
also exploring ways for AI to transact directly on consum- behavior.
ers' behalf. The infrastructure is developing quickly, but
the rules governing disputes are struggling to keep pace. The result is an increase in false declines: authorized

A fourth category of dispute transactions rejected by systems attempting to prevent
fraud. False declines do not generate chargebacks, but the
For most of the history of card payments, disputes have commercial consequences are immediate. Merchants lose
broadly fallen into three categories: criminal fraud, mer- revenue, damage customer trust and risk becoming invis-
chant error and buyer's remorse. Each has established ible to AI agents that learn over time which merchants
rules, evidence requirements and liability paths. consistently reject transactions.
18