AI agents and security: Building trust in the age of agentic commerce

Making decisions quickly and safely

In some global payments test pilots, AI agents have been responsible for full payment flows within pre-set parameters. For example the CEO of a freight broker company interviewed by Paul Berger for a Wall Street Journal article stated that his organization is beginning to use agentic capabilities for global forwarding activities, and he’s had agents deliver a quote in about an hour rather than the several days it previously took.¹ Assumedly, the payment itself can process just as quick. That kind of speed is staggering when you consider the baseline. In traditional B2B cross-border payments, settlement can take two to five days. In agentic pilots, those same payments can be completed in well under five minutes.

The efficiency gains are impressive, but there is, of course, a flip side: the window for detecting fraud or preventing an error is compressed. Where fraud detection systems used to have hours or days to identify and stop a problem, businesses now have minutes if not seconds. This fundamentally changes the risk calculus for everyone involved.

Trust and accountability become non-negotiable

The efficiency gains are impressive, but there is, of course, a flip side: the window for detecting fraud or preventing an error is compressed. Where fraud detection systems used to have hours or days to identify and stop a problem, businesses now have minutes if not seconds. This fundamentally changes the risk calculus for everyone involved. It’s clear the addition of AI agents in B2B payments require new trust frameworks.

With agentic payments, trust moves from being a background consideration to a front-line requirement. Before any transaction occurs, AI agents must be identified and authenticated with far more rigor than a simple login or API key. It is not enough to know that a request came from the right software; there needs to be confirmation that the agent has the correct request and parameters for it, the proper permissions and is acting on behalf of a legitimate, verified business.

During the transaction, the agent’s behavior must align exactly with the buyer’s intent. On some occasions, that may mean requesting human review at key decision points; on every occasion, it means not bypassing negotiated supplier terms, not missing bundled offers or discounts and not selecting payment rails that are slower or riskier than necessary.

The most advanced agentic pilots already use APIs to confirm intent, ensuring an agent cannot simply “decide” to spend without an auditable record of human approval. Trust, after all, is built not only on security controls but also on user choice and transparency. This is why embedded transaction controls and continuous authentication are crucial for agentic payments.

After agentic transactions are completed, accountability is non-negotiable. Financial institutions and their business customers should be able to trace every decision point. Immutable audit trails can show exactly which signal triggered a payment, what terms were applied, which loyalty benefits or discounts were used and which payment rail was chosen. This level of transparency not only enables rapid dispute resolution but also helps refine agent behavior over time, reducing errors and preventing repeat issues.

Agentic acceptance is a far easier lift than businesses may expect

The good news is that the foundational tools to make agentic transactions secure already exist. The same identity frameworks, global acceptance networks and transaction controls that help secure payments today can be extended to secure agentic transactions. Tokenization, for example, can help ensure sensitive credentials are never exposed. Fraud monitoring can flag anomalies in milliseconds. Built-in dispute resolution processes can be triggered automatically when conditions are not met. Agentic acceptance is, in that sense, a far easier lift than businesses may expect. There’s the old saw about how everyone wants things good, fast and cheap, but they need to pick two of the three. Agentic commerce may obviate the need for businesses to choose.

Quick agentic settlement means businesses can move funds outside of traditional banking hours. When combined with embedded safeguards, programmable payments and intelligent rail selection, agentic commerce can allow companies to align payment execution with working capital needs, supplier preferences and internal compliance requirements.

Banks are positioned to lead the agentic revolution

Agentic commerce doesn’t just benefit large enterprises; the same frameworks can serve small and medium-sized businesses (SMBs), independent creators and gig workers. For example, a verified AI agent embedded in a marketplace platform in the U.S. could trigger instant payouts to a freelancer in India, a just-in-time supplier payment in France and a milestone-based release of funds to a subcontractor in Kenya, all in ways that meet company compliance requests and are locally optimized, and fully traceable.

For banks, the opportunity with agentic commerce is clear. As longtime custodians of trust in commerce, their role is more critical than ever. By extending their trust infrastructure to cover AI participants, they can stay relevant in a future where transactions are increasingly initiated by code rather than people.

Thriving in an agentic ecosystem means co-creating solutions with technology providers and fintechs, and integrating with APIs, ERP platforms and monitoring systems that can certify and supervise agents in real time. It also means participating in the development of industry-wide standards for agent identity, hashing, permissions and transaction logging.

Conclusion

The financial institutions that drive agentic commerce forward will not just be protecting their customers: they will be actively shaping the future of B2B commerce. The leaders will be the ones enabling faster, safer, more transparent value movement across borders and industries. And they will be the ones ensuring that as commerce speeds up, trust keeps pace.