DBS Bank to let AI agents make payments on behalf of customers

Artificial intelligence in banking has long remained in the role of a smart advisor: analyzing expenses, suggesting investments, detecting suspicious transactions. But Singapore's DBS Bank, Southeast Asia's largest financial institution, decided to go further and give AI agents the right not just to advise, but to act. In partnership with Visa, the bank launched a pilot project in which artificial intelligence can independently make payments on behalf of customers.

At the heart of the pilot is the Visa Intelligent Commerce platform — an infrastructure that the payment giant is developing specifically for the era of autonomous AI agents. The idea is that the user delegates a specific task to the agent — for example, book a hotel, pay a subscription, or buy a product — and the agent executes it, including the final payment stage, without needing to return to the user for confirmation at each step. This fundamentally differs from existing voice assistants and chatbots, which can suggest options but ultimately hand control back to the user at the payment stage.

Technically, the system relies on tokenization of payment data. Instead of giving the AI agent real credit card data, Visa generates special tokens with limited validity periods and predetermined limits. The agent receives permission to conduct a transaction within strictly defined parameters: maximum amount, purchase category, time window. If the request exceeds these boundaries, the system requires explicit confirmation from the customer. This approach is designed to minimize fraud and unauthorized charges risks, although it is clear that as such systems scale, new attack vectors will emerge.

DBS's choice as a partner for the pilot is no accident. The bank consistently positions itself as a technology leader in the Asia-Pacific region. It was among the first in the world to launch a full-fledged digital banking platform, actively invests in cloud technologies and machine learning, and its internal AI infrastructure processes millions of transactions daily. For Visa, this pilot is an opportunity to test Intelligent Commerce under real Asian market conditions, where mobile payments and digital wallets have long been the norm, and consumers are significantly more open to financial technology innovations than, say, in Europe.

But behind the technological elegance of this project lies a whole layer of questions that the industry does not yet have answers to. The main one is responsibility. If an AI agent makes an erroneous purchase by misinterpreting the user's request, who bears the responsibility? The bank that issued the token? Visa, which provided the infrastructure? The agent developer? The user himself, who delegated authority? Existing legislation in most jurisdictions simply does not account for situations where an autonomous program controls other people's money. Regulators in Singapore, traditionally progressive on fintech issues, will likely be among the first to have to establish legal frameworks for such scenarios.

It is also worth noting the competitive context. Visa is not the only company preparing payment infrastructure for AI agents. Mastercard is developing its own Agent Pay program, tech giants like Google and Apple are integrating payment capabilities into their AI ecosystems, and agentic AI startups are attracting record investments. The race to become the "payment rail" for autonomous agents is just beginning, and the stakes are exceptionally high: according to various forecasts, by 2030 the volume of transactions initiated by AI agents could be measured in trillions of dollars.

The DBS and Visa pilot is not simply an experiment by one bank. It is a signal of a fundamental shift in the very concept of financial services. Until now, banks have built their products around human interaction: interfaces, applications, notifications — everything was designed with the assumption that humans make decisions. Now, a parallel channel is emerging where the client of the bank effectively becomes a program. And if this pilot demonstrates the viability of the model, we will see how the world's largest banks begin to restructure their architecture for a new "user" — an AI agent acting on behalf of a human but making decisions independently.