Page 38 - gs260701
P. 38
Insights and Expertise
This creates an entirely new layer of uncertainty for dis-
pute management.
Five questions your merchants should answer
before deploying AI agents The situation becomes even more complicated when mul-
tiple AI agents begin interacting with each other. One AI
may represent the consumer while another represents
As excitement around agentic commerce grows, the merchant. Both systems may negotiate prices, deliv-
many businesses are focusing on what AI agents ery terms, subscriptions or recurring commitments before
can do. Merchants should also consider what AI payment is initiated. If a misunderstanding occurs be-
agents should be allowed to do. Before introduc- tween two autonomous systems, determining responsibil-
ing AI into purchasing or payments, merchants ity becomes significantly more difficult than identifying
should ask five practical questions: which human made a mistake.
• Who is authorized? Can the AI complete
purchases independently, or should it re- New controls
quire human approval above certain dollar Payment providers and acquiring banks will inevitably
amounts? face pressure to introduce new controls. They will need
• What are the limits? Should spending to determine the level of authority an AI agent possess-
caps, approved vendors, product categories es, and also how instructions should be verified after the
or time-of-day restrictions be established? transaction takes place. The industry may eventually re-
• How will fraud be detected? Existing quire completely new audit trails that capture not only the
fraud tools focus on human behavior. Mer- payment itself but also the reasoning process that led to
chants should evaluate whether they can the payment decision.
recognize unusual AI-driven activity or
manipulated product information. Regulators will face similar challenges. Existing payment
regulations assume that decision-making originates from
• Who is accountable? If an AI agent makes an identifiable person or organization, but agentic pay-
an unauthorized purchase or commits to ments introduce a third participant into the relationship.
an unwanted subscription, who investi-
gates the problem, manages disputes and The AI is neither a customer nor a regulated financial in-
absorbs the financial loss? stitution, but it can easily influence decisions that move
• Is there an audit trail? Businesses should significant amounts of money. This creates difficult ques-
be able to reconstruct not only what pay- tions regarding accountability, consumer protection,
ment occurred but why the AI reached its transparency and oversight.
decision. Fraud
As agentic commerce matures, merchants that
establish governance before deployment will likely Today's fraud monitoring systems are designed around
adapt more easily than those that treat AI solely as human behavior: analyzing spending patterns, locations,
another technology upgrade. The goal is not simply to devices and transaction history to determine whether a
automate purchasing—it is to ensure that automation payment appears legitimate.
operates within clearly defined business, compliance
and risk management frameworks. Agentic payments change this dynamic entirely. AI agents
may search hundreds of websites, interact with multiple
merchants simultaneously and complete transactions at a
Responsibility speed no human can match. Fraudsters will quickly adapt
to this new environment, and they will manipulate AI
One of the biggest unanswered questions in agentic pay- agents through fake websites, misleading product infor-
ments is responsibility. The entire payment ecosystem, mation, fraudulent offers or artificially generated reviews
from card schemes and banks to payment providers and designed to influence purchasing decisions.
regulators, was designed around a simple assumption: a
human makes a decision, authorizes a transaction and ac- An AI agent may identify the cheapest option available,
cepts the consequences of that decision. Agentic payments but that does not mean it has identified the safest one. As a
challenge that assumption entirely. result, fraud prevention will no longer focus only on veri-
fying the identity of the payer, but also increasingly on
When an AI agent purchases the wrong product, who is verifying whether the information and decisions used by
responsible? If it misunderstands the user's instructions the AI can be trusted.
and spends more than intended, who carries the financial
loss? And what about manipulation, fraud or misleading You're not ready for agentic AI
information? Most businesses do not even have a payment and bank-
ing strategy today. They work with various payment pro-
38
