Agentic Commerce Isn’t an AI Problem for Banks. It’s an Authorization Problem
By Kathleen Peters at Experian North America
Simple Subscribe
Subscribe Now!
The debate around agentic commerce has largely centered on whether AI agents are reliable enough to shop, negotiate and make purchases on behalf of consumers. Reliability matters, but that isn’t the question banks should be asking.
The more important question is this: Who will decide what those AI agents are allowed to do, and under what conditions?
Banks have spent decades building the infrastructure that allows consumers to move money safely. But in an economy where AI agents increasingly act on behalf of people, the competitive advantage may shift from processing payments to defining permissions.
Key insight: Banks don’t have an AI problem. They have an authorization problem.
Need to Know:
- Agentic commerce creates a new challenge for banks: verifying who is behind a transaction and what an AI agent has permission to do on that person’s behalf.
- Banks can build on existing identity verification, fraud detection and behavioral analytics while developing new ways to verify agents and govern delegated authority.
- Financial institutions can start taking steps today to give consumers meaningful control over AI agents and help build trust across the broader commerce ecosystem.
Permission is the Next Battleground
For years, innovation in financial services has focused on making transactions faster, more seamless and more secure. We’ve optimized payment rails, modernized digital banking, reduced friction at nearly every point in the customer journey, and established clear expectations for accountability and remediation when something goes wrong.
Whether the issue is accidental, such as ordering the wrong size, or malicious, such as fraud or abuse, all parties understand how it is resolved and who is responsible.
What’s changing: Agentic commerce throws a wrench into this balance.
When an AI agent purchases an airline ticket or reorders inventory for a small business while its owner sleeps, the question of whether that agent had the authority to act is just as important as the fidelity of the payment itself.
Consumers and businesses need confidence that an autonomous system acted within clearly defined boundaries, using permissions they established and can modify at any time.
Read more: When an AI Agent Makes an Incorrect Purchase, Who’s Responsible? Amex Offers a Partial Answer
A New Era of Delegated Authority
For decades, authentication has largely answered a single question: Who are you?
Agentic commerce introduces two new, equally important questions: What is this agent authorized to do, and what is your intent?
The gap between these questions and their answers is where risk lives.
Consider a small business owner whose AI agent is tasked with reordering supplies. The agent may have legitimate credentials and access to a real payment method. But without the right guardrails, it could place an order that exceeds the intended budget, buy from a vendor the owner would never have approved, or execute a transaction after the business relationship has been discontinued.
Banks already understand this distinction in other contexts. Credit limits, merchant controls, transaction alerts and fraud detection all establish boundaries around financial activity. What’s changing is the actor initiating the transaction.
Key insight: When software acts instead of people, those boundaries need to be explicit, dynamic and governed, not assumed.
Read more: Who Will Protect Banking Consumers’ Rights in the Age of AI?
Banks Face a New Competitive Threat
Fraud, hallucinations and AI safety are real concerns. But they are often downstream consequences of a larger strategic question:
Who owns agent authorization?
Banks have traditionally occupied that ownership role because they sit at the center of trusted financial relationships.
However, technology platforms, payment networks, digital wallets and AI providers all have incentives to become the layer through which consumers grant permission for autonomous transactions.
Competitive insight: If that happens, banks risk becoming the infrastructure underneath the customer relationship rather than the institution defining it.
The history of digital commerce is filled with examples of companies that assumed they would naturally retain strategic control, only to discover that another layer of the ecosystem had become more valuable.
Banks should not assume this moment will be different.
Read more: ‘Know Your Agent’ is a Must When Autonomous Payments Can Be Fraudsters’ Entry Point
What Banks Can Do Now to Prepare
It’s tempting to think success in agentic commerce will be determined by who builds the smartest AI or agentic protocol.
I believe that’s the wrong rubric. Consumers don’t need AI that is infinitely more intelligent before they’ll trust it to make routine purchases. They need AI that is governable.
What an agent can purchase, how much it can spend, which merchants it can interact with, how those permissions can be revoked or modified, and who is accountable when something goes wrong — these are trust and authorization problems. Banks have been solving versions of them for decades.
How to move forward: Banks don’t need to wait for agentic commerce to reach maturity before preparing for it. They can start by extending principles they already use to protect consumers into a world where software increasingly acts on their behalf.
A plan of action:
1. Banks need to evolve the “know your customer” model to account for a new participant in the transaction: the agent.
Establish the link between the human, the agent, and the intent behind the transaction. Knowing that valid payment credentials were used won’t be enough; banks and merchants will increasingly need to know whether an agent is connected to a verified person, whether that agent itself can be trusted, and whether it had permission to initiate a particular transaction.
2. Build on the fraud infrastructure that already works.
Banks have decades of investment in identity verification, fraud detection and behavioral analytics at their disposal. Agent verification can become another signal within those systems. Over time, banks can establish a baseline for how a particular agent normally behaves and use anomaly detection to flag activity that falls outside it.
An agent that suddenly begins transacting differently than expected should raise questions — just as unusual human behavior does today.
3. Give consumers meaningful control over delegated authority.
For example, someone may be comfortable letting an agent automatically make small routine purchases, but unwilling to let it buy a new car or a plane ticket.
Financial institutions should anticipate that spectrum by developing ways for consumers to understand what authority they have delegated and establish boundaries around what an agent can do. Just as important, consumers need confidence that when something does go wrong, there is a clear process for making it right.
4. Don’t try to solve the issue of agentic trust alone.
Banks, merchants, payment networks, technology platforms and AI providers each see only part of the transaction. Establishing trust will require these players to work together around interoperable standards and approaches rather than locking consumers into a single AI platform or payment ecosystem.
The technology will continue to change quickly. So the trust layer needs to work across the channels and platforms consumers choose.
Read more: Your Customers Will Blame You When Their Shopping Bots Go Rogue
Success Hangs on Permission
From ATMs to online banking to mobile payments, the banking industry has successfully navigated major technological transitions before. Each of these innovations required new trust frameworks.
Agentic commerce will be no different.
What to watch: For the past two decades, banks have competed to become consumers’ preferred payment provider. Over the next decade, they may compete to become consumers’ preferred authorization authority.
Banks that begin building robust models for delegated authority, dynamic authorization and consumer-controlled permissions will not only reduce fraud risk but also help define how autonomous commerce operates.
Bottom line: Financial institutions that wait may discover that others have already established the rules.
Read next: How — and Why — Your Bank Should Be Prepping for Agentic Payments Now
