Keeping the faith: humans still belong in payments

Let’s start with a working definition for agentic commerce: “a new form of online and mobile shopping, in which an AI agent ‘closes the loop’ or completes tasks for a user – such as searching for items, comparing options, and making a purchase – with limited or no manual inputs needed from that user.”¹

Consider what the term “limited or no manual inputs” means. Proponents maintain agentic commerce moves the needle from efficiency to autonomy within the transaction ecosystem (a good thing, in their view). Critics beg to differ, citing – among others – issues with the current technologies, insufficient regulation over guardrails and the like, and a lack of education across the ecosystem. 

“Three industry-wide trends converged in 2026,” noted Ivan Vukelikj, Payment Security Expert at G+D Netcetera. “First, agentic commerce started moving from the concept stage to early deployment, with several payment service providers (PSPs) publishing concrete frameworks for AI agents that transact on behalf of users. Second, European regulations sharpened liability and consent requirements while reinforcing strong customer authentication (SCA): this really highlighted the question of how consent works when an agent is acting for a person. Third, fraud losses on instant payments and account-to-account (A2A) rails continued to climb, driven by manipulation of the payer rather than a failure of authentication. This exposes a trust problem that automation alone cannot solve.”

Much of the discourse around payments focuses on reducing friction. But in a world where payments are run by layers of orchestration, and increasingly AI, where do humans belong? 

Put another way, consumers and merchants want “frictionless” instant payments. But everyone involved – including banks, PSPs, and other stakeholders – also expect rigorously secure transactions. Add the question of establishing liability when an agent is involved, and the complexity of the situation emerges. 

Call it the trust paradox, or something else entirely. Resolving it – and establishing trust across the chain – is key to the scalable growth of agentic commerce.

A new kind of commerce

Consider an agentic transaction. You indicate to your agent that you’d like a pair of shoes in a certain color, under a certain price point, delivered to your door by a certain date. The agent has done the product discovery, examined the options, and executed the transaction. The “click-to-buy” moment doesn’t occur anymore for the user themselves. 

This has implications for every step in the payment process, from initiation to authorization to settlement. While theoretically close to frictionless, it introduces points of failure that require addressing. To take just one instance: perhaps you changed your mind about the shoes but forgot to let your agent know, and it went ahead and ordered them anyway. “If a charge is made that a user disputes, who is liable?” asked Vukelikj. “Is it the user, the merchant, the issuer? Is it the agent’s vendor?”   

Further, who does the user turn to for recourse when the entire process is geared toward automation? An AI agent ordered the shoes, after all: will speaking to a bot find a solution? And when you want to escalate, whom do you escalate to?

A concerted initiative

The good news is that the payment industry – and the larger tech ecosystem – is aware of the scale of the issue and is working at speed to deal with it. Working groups across consortiums and associations such as W3C, the FIDO Alliance, and EMVCo are collaborating to establish frameworks on topics such as liability across use cases in agentic commerce. This includes everything from high-level policy to nuts-and-bolts processes that deal with authorization and authentication. 

Among other things stakeholders are looking at: 

1. A liability framework that can integrate new players, including AI platforms
2. Interoperability across the ecosystem with standards for all players and use cases
3. Robust security with global acceptance, so PSPs, issuers, and merchants can communicate directly and easily

“We’re talking about technologies that weren’t even in the public domain more than three years ago,” said Vukelikj. “With AI, the frictions that currently exist between the digital and “human” worlds are being erased. Payments are moving from being human-driven to orchestrated and agent-led flows. But that is the space where regulation must step in. All entities in the ecosystem need to be properly integrated and protected. The end user needs to know their data will be safeguarded and not used for fraudulent purposes.”

The changing face of fraud

According to a recent European Central Bank (ECB) study, SCA remains highly effective against the sorts of fraud that it is designed to counter, especially card payments. However, “new types of fraud are on the rise, particularly the manipulation of players, which will require additional and new mitigation approaches.”² The United Kingdom has reported similar trends, with authorized push payment (APP) fraud losses up 12% from the year before.³

APP frauds are when fraudsters trick their mark into willingly transferring money into an account or other vehicle they control, whether through impersonating someone else, offering an investment that doesn’t exist, or some other trick. The confluence of technical fraud and social engineering is a major issue for everyone involved. 

Further, fraud detection in the era of agentic commerce will become more complex, as fraud detection models are based in part upon patterns of human behavior. Agents aren’t human. They behave differently. At one end of the scale, this could trigger rejections of perfectly legal and user-intended transactions. At the other end, fraudulent transactions might escape notice. 

Which all suggests that even as the payments cycle moves to greater automation and even autonomy, keeping people involved is ever more critical – in initiating a transaction, for instance, or confirming it, and intervening when required.

Keeping payments human

“The most successful invisible payment systems – those that happen in the background, without an explicit ‘check out’ – have the clearest, most efficiently placed human checkpoints,” stated Vukelikj. “Invisibility scales convenience. But human agency scales trust.”

What that means in practice is moving beyond thinking of the bits that make the process work as “friction.” This includes tokenization, authentication, and managing identity components in the most secure way. In a sense, this is the visible architecture of human agency in an otherwise highly automated payments flow that is over in seconds. 

“This is where the user confirms their choices,” stated Vukelikj. “This is where they delegate their details and intervene – or request an intervention – when they see something out of the ordinary.” The tech is moving at speed, as it always does. But it won’t scale in terms of usage and adoption until people trust it.

What that means for the industry – including issuers, acquirers, and merchants – is to ensure people stay in the payments flow in a meaningful way. This would mitigate a flood of disputes and reversals as the agentic ecosystem gathers pace.

“It requires education,” noted Vukelikj. Not just of users, but merchants and others in the ecosystem. The tech and regulatory landscapes will change dynamically. What helps everyone is when the architecture is built to exacting specifications by a trusted partner with established credentials in payments and identity technology. 

G+D Netcetera is such a partner. It provides the trusted integration and security layer that brings together authentication, tokenization, CIAM, program frameworks, and intent controls. This enables orchestrated and automated payments, while keeping humans in control of the decisions that matter, thus strengthening trust across the ecosystem. 

“Perhaps one way to resolve the trust paradox is to accept what seems like a contradiction as well,” smiled Vukelikj. “Invisible payments will truly take off once humans are reassured that they will retain their visible agency in that ecosystem.”

Published: 15/09/2026