Agentic commerce has a payments problem hiding in plain sight. The financial system does not yet have a universal way to authenticate a machine’s authority to make a decision.
And time is running out. Artificial intelligence agents can already shop, negotiate and transact on behalf of humans.
“The volume is small,” Troy Mann, senior vice president, product management at FIS Global, told PYMNTS.
But volume is volume. Consumers who have stored payment credentials in wallets or online commerce environments are already entering an ecosystem being rebuilt for agentic transactions.
That means banks may eventually need to recognize not simply that a transaction came from an AI agent, but which agent it was and what it was entitled to do.
“Intent has to be expressed and captured,” Mann said.
Agentic Commerce Is Here, but the Permission Layer Is Still Catching Up
Payment systems were built largely around identifying accounts, credentials and transactions. Autonomous commerce adds the software making the decision, another object that needs to be identified.
Today, most agentic commerce still begins with human intent, Mann said. A buyer specifies what they want, potentially including a price ceiling, shipping requirements or other constraints. An agent can search within those parameters, but ultimately the system needs evidence that the buyer authorized the resulting transaction.
Banks therefore need to evaluate more than whether a credential is legitimate. They need to know which agent is using it, what that agent was authorized to do and whether the transaction matches the original mandate.
Tokenization is a “major layer” of that security architecture, Mann said. But tokenizing credentials solves only part of the problem.
Merchants may know what the customer requested. AI platforms may know the instructions supplied to the agent. Banks know the account, payment credential and transaction history. These signals are not necessarily traveling together. The emerging challenge is that much of the information needed to establish agentic legitimacy sits outside the issuer’s traditional field of vision.
A Valid Credential Is No Longer Enough; Audit Trails Become Part of the Payment
The issue with endemic data fragmentation is that agentic transactions introduce questions conventional fraud systems were not designed to answer. Did the agent authenticate correctly? Was it permitted to store the credential? Which AI agent initiated the transaction? Does the purchase conform to the authority the customer originally granted?
FIS is working with financial institutions to identify agentic transactions, apply different fraud scoring and reporting, and incorporate signals across the payment life cycle. The industry ultimately needs richer information flowing between merchants, agents, networks and issuers, Mann said.
“Data is the key to everything,” he said.
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
That creates an unusual near-term conclusion for an industry that has spent decades removing friction from checkout. Agentic commerce may initially need to put some back.
“A little bit of friction isn’t a bad thing,” Mann said, adding that one possibility is virtual cards specifically designated for agentic commerce.
Businesses could restrict them by transaction amount, merchant, time window or other parameters, creating a credential whose permissions mirror the authority granted to the AI. Other approaches could make security credentials dynamic for agent-initiated transactions. The point is not simply to make fraud harder. It is to make machine authority observable.
For now, Mann recommended a “crawl, walk, run” approach in which humans remain involved in establishing intent and authorization.
Trust Is Becoming Infrastructure
Agentic commerce ultimately poses a broader architectural problem. Banks may eventually need to recognize not simply that a transaction came from an AI agent, but which agent it was and what it was entitled to do.
For businesses, the immediate opportunity is therefore inseparable from the control architecture surrounding it. Merchants want agents that can convert customer intent into purchases. Banks want to remain top of wallet. Buyers want automation without surrendering control of their money.
Making all three possible requires permission, authentication, tokenization, data sharing and auditability to operate as one system.
Mann said he remains optimistic about where that leads. Agentic commerce is “going to change people’s lives … for the better, if it’s done smartly and if it’s done with transparency and trust.”
The race toward autonomous commerce may therefore depend less on how intelligent the agents become than on whether the financial system can prove what they were allowed to do.
Watch the full PYMNTS interview with FIS Senior Vice President Troy Mann to hear more about:
- Why AI agents can already spend your money, even if banks aren’t quite ready. Agentic commerce is live at small volumes, forcing banks to verify not just the payment credential, but the agent, its permissions and the buyer’s original intent, Mann said.
- Why payments may need more friction, not less. Virtual cards, spending limits and time-based controls could give AI agents tightly defined purchasing power while banks build stronger safeguards, Mann said.
- Why AI payments have a data problem. Crucial signals about intent, authorization and transactions remain split among merchants, agents and banks, leaving data sharing and auditability as prerequisites for greater autonomy, Mann said.
For all PYMNTS AI coverage, subscribe to the daily AI newsletter.