Banco Santander and Mastercard have completed what they describe as Europe’s first live, end-to-end payment executed by an artificial intelligence agent. The transaction was not conducted in a sandbox or test environment — it ran on Santander’s live payments infrastructure, using real banking rails and real money, marking a significant milestone in the transition of agentic commerce from concept to production.
The pilot used Mastercard Agent Pay, an agentic payments protocol that enables AI agents to initiate and complete transactions on behalf of consumers within predefined permissions. PayOS handled the end-to-end orchestration. The result: an AI agent discovered a product, selected it, initiated payment, and completed the transaction — all without direct human intervention at the point of sale, while maintaining the security, governance, and consumer protection standards required under a regulated banking framework.
The announcement is significant not because an AI agent made a payment — that has been demonstrated in controlled environments for months — but because it happened on live banking infrastructure at one of the world’s largest financial institutions. With approximately €1.84 trillion in total assets, Santander’s involvement signals that agentic commerce is moving from fintech experimentation into mainstream banking.
What Actually Happened in the Pilot
The technical architecture of the transaction involves three distinct layers. Mastercard Agent Pay provides the payment protocol — a framework that registers and verifies AI agents, issues tokenised payment credentials, and ensures that every agent-initiated transaction operates within permissions defined by the consumer. The protocol builds on Mastercard’s existing tokenisation technology, the same infrastructure that secures mobile contactless payments and card-on-file transactions globally, extended to support autonomous AI agents as a new transaction initiator.
Santander contributed the live banking infrastructure — the actual payment rails, account systems, and regulatory compliance layer that processed the transaction. This is the element that distinguishes the pilot from previous agentic commerce demonstrations: the payment was not simulated or routed through a testing environment but processed through the same systems that handle Santander’s day-to-day transaction volume across Europe.
PayOS served as the orchestration layer, connecting the AI agent to the merchant, the payment protocol, and the banking infrastructure. This middleware function is critical because agentic commerce requires coordination between systems that were not originally designed to interact — conversational AI platforms, payment networks, merchant checkout systems, and bank processing engines all need to communicate in real time for an agent-initiated transaction to complete successfully.
The Mastercard Agent Pay Framework
Mastercard launched its Agentic Payments Programme in April 2025, introducing a framework designed to bring the same trust and security standards that govern traditional card payments into the emerging world of AI-initiated transactions. The core innovation is Mastercard Agentic Tokens — dynamic digital credentials that allow AI agents to transact on behalf of consumers while maintaining full authentication, permission boundaries, and transaction visibility.
The system works by requiring AI agents to be registered and verified before they can initiate payments. Each agent receives tokenised credentials tied to a specific consumer’s account, with predefined spending limits, merchant categories, and transaction types that the consumer has explicitly authorised. Every transaction initiated by an agent is traceable — the consumer, the issuing bank, and the merchant can all identify which agent acted, on whose behalf, and within what parameters.
Mastercard has been scaling the ecosystem rapidly. Citi and US Bank cardholders were the first to be enabled for Agent Pay-powered transactions, with all US Mastercard cardholders expected to be enabled by the 2025 holiday season and global rollout following shortly after. The network has integrated Agent Pay with PayPal’s wallet, giving hundreds of millions of consumers access to agentic commerce. It has also partnered with Stripe, Google, and Ant International’s Antom to make secure agentic transactions accessible to digital merchants globally. Live agentic transactions have now been completed in the US, Australia, New Zealand, Singapore, Malaysia, India, and South Korea — with the Santander pilot adding Europe to the list.
Why This Matters for Open Banking and Payment Infrastructure
The Santander-Mastercard pilot sits at the intersection of two major infrastructure trends that this publication covers closely: the evolution of open banking APIs and the emergence of AI agents as autonomous participants in financial transactions.
Open banking, as defined by PSD2 and the forthcoming PSD3/PSR framework, established the principle that third-party providers can access bank account data and initiate payments on behalf of consumers through standardised APIs. Agentic commerce extends this principle by adding a new category of third-party actor: the AI agent. Instead of a human consumer clicking through an online checkout, or a payment initiation service provider triggering a bank transfer, an AI agent autonomously discovers products, selects options, initiates payment, and completes the transaction — all within permissions that the consumer has defined in advance.
This raises a set of infrastructure questions that the payments industry is only beginning to address. How should AI agents be authenticated? How do existing Strong Customer Authentication (SCA) requirements apply when the transaction initiator is software rather than a human? How should consent be managed when an agent operates autonomously over time rather than receiving explicit approval for each individual transaction? How do merchants distinguish legitimate AI agents from automated bots attempting fraud?
Mastercard’s approach — registering agents, issuing purpose-specific tokens, and maintaining full transaction traceability — represents one answer to these questions. Visa has taken a parallel approach with its Trusted Agent Protocol, developed in collaboration with Cloudflare, which establishes a framework for secure communication between AI agents and merchants. The fact that both major card networks are investing heavily in agentic payment infrastructure suggests that the industry expects AI-initiated transactions to become a significant share of commerce volume within the next few years.
Santander’s AI Strategy: From ChatGPT Enterprise to Agentic Payments
Santander’s participation in the pilot reflects a broader AI strategy that has been building momentum over the past year. The bank’s AI initiatives are led by chief data and AI officer Ricardo Martín Manjón, who was hired from BBVA in March 2025. Under Manjón’s leadership, Santander formed a strategic partnership with OpenAI in August 2025, connecting up to 30,000 bank employees with ChatGPT Enterprise in what was described as one of the fastest deployments of its kind in the European banking sector.
The progression from internal AI deployment to live agentic payments is notable. Most banks that have adopted AI tools have focused on internal use cases — document processing, customer service augmentation, code generation, data analysis. Santander’s agentic payment pilot moves the AI deployment perimeter outward, into customer-facing transaction infrastructure where the stakes are materially higher. A chatbot that generates an imperfect internal summary is an inconvenience. An AI agent that executes an incorrect payment on live banking rails is a regulatory and reputational event.
The bank has confirmed it will move into extended testing and scaling, exploring additional use cases and partnerships while maintaining regulatory alignment. This measured language is significant — it signals that the pilot was a proof of concept, not a commercial launch, and that considerable work remains before agentic payments become a standard feature of Santander’s customer-facing product suite.
The Competitive Landscape: A Race to Define Agentic Commerce Standards
The Santander-Mastercard pilot does not exist in isolation. The past twelve months have seen an acceleration of agentic commerce activity across the payments industry. Visa launched its Trusted Agent Protocol in October 2025 and reported hundreds of secure agent-initiated transactions by December. DBS Bank in Singapore became the first issuer to pilot Visa Intelligent Commerce for everyday payments in February 2026. Mastercard showcased the first fully authenticated agentic commerce transaction in India using cards issued by Axis Bank and RBL Bank. South Korea’s first live agentic transaction was completed in March 2026 through the Mastercard Agent Pay framework.
The underlying competition is not just between banks or between card networks — it is a contest to define the standards and protocols that will govern how AI agents interact with the payments ecosystem. Mastercard is working with Google’s Universal Commerce Protocol, OpenAI’s Agentic Commerce Protocol, and other interoperability frameworks. Visa is building its own protocol stack. Meanwhile, technology companies like Google, OpenAI, and Microsoft are developing their own agent-to-merchant and agent-to-agent communication standards.
Whoever establishes the dominant protocol framework for agentic payments will hold a position of considerable influence over the next generation of digital commerce. The card networks are leveraging their existing trust infrastructure — tokenisation, fraud detection, dispute resolution — to argue that agentic payments should flow through established payment rails rather than through new, unregulated channels. This is a strategic play as much as a technical one: if AI agents transact primarily through card network infrastructure, Mastercard and Visa maintain their position as the central intermediaries of global commerce even as the transaction initiator shifts from human to machine.
What Comes Next
The Santander-Mastercard pilot is a milestone, but it is the beginning of a long integration process rather than a finished product. Several critical questions remain unresolved. Regulatory frameworks for agentic payments are still being developed — the European Banking Authority has not yet issued specific guidance on how existing payment services regulations apply to AI-initiated transactions. Consumer protection frameworks need to address scenarios where an AI agent makes a purchase that the consumer disputes. Liability allocation for agent errors — who is responsible when an AI agent exceeds its permissions or makes a suboptimal purchasing decision — remains an open legal question.
For banks, the opportunity is clear but the execution is complex. Agentic commerce represents a potential new distribution channel for payment products, a way to increase transaction volume, and an opportunity to deepen customer engagement through AI-assisted financial services. But it also requires banks to extend their infrastructure to support a new category of transaction initiator, to build or integrate agent management capabilities, and to adapt their compliance and risk frameworks to accommodate autonomous AI actors operating on live banking rails.
The financial institutions and payment networks that get this right will define how commerce works in the age of AI agents. Those that move too slowly risk being disintermediated by technology companies that build their own payment infrastructure. Those that move too fast risk the regulatory and reputational consequences of deploying AI in production financial systems before the governance frameworks are mature. The Santander-Mastercard pilot suggests that the smartest approach may be exactly what these two companies are demonstrating: controlled experiments on live infrastructure, conducted in close dialogue with regulators, with the ambition to scale but the discipline to test first.
Frequently Asked Questions
What is an agentic payment?
An agentic payment is a transaction initiated and completed by an AI agent on behalf of a consumer, without the consumer manually interacting with a checkout process. The AI agent discovers products or services, selects options based on the consumer’s preferences, initiates the payment using tokenised credentials, and completes the purchase — all within permissions that the consumer has defined in advance. The consumer retains control through predefined spending limits, merchant categories, and the ability to revoke agent permissions at any time.
What is Mastercard Agent Pay?
Mastercard Agent Pay is a payment protocol launched in April 2025 that enables AI agents to initiate and complete transactions securely on behalf of consumers. The system uses Agentic Tokens — dynamic digital credentials built on Mastercard’s existing tokenisation technology — to authenticate AI agents, enforce spending permissions, and maintain full transaction traceability. The protocol integrates with Microsoft Azure OpenAI Service and Microsoft Copilot Studio, and has been adopted by banks including Citi, US Bank, Westpac, and now Santander.
Why is the Santander pilot significant?
Previous agentic payment demonstrations were conducted in sandbox or test environments. The Santander pilot is significant because it was executed on the bank’s live payments infrastructure — real banking rails processing a real transaction. This demonstrates that agentic payments can operate within the security, governance, and regulatory compliance standards of a major European bank, moving the technology from proof-of-concept to production-grade viability.
How do agentic payments relate to open banking?
Open banking established the regulatory and technical framework for third-party providers to access bank account data and initiate payments through APIs with consumer consent. Agentic commerce extends this model by introducing AI agents as a new category of third-party actor that can autonomously initiate transactions. Both rely on the same foundational principles — consumer consent, secure API access, regulatory oversight — but agentic payments add complexity around authentication, autonomous permission management, and the governance of non-human transaction initiators.
What is the difference between Mastercard Agent Pay and Visa’s Trusted Agent Protocol?
Both are frameworks designed to enable AI agents to initiate secure payments, but they take different approaches. Mastercard Agent Pay focuses on extending its existing tokenisation infrastructure to issue agent-specific credentials, with integration points into Microsoft, PayPal, Stripe, and Google ecosystems. Visa’s Trusted Agent Protocol, developed with Cloudflare, establishes a framework for secure communication between AI agents and merchants, focusing on verifying agent identity and distinguishing legitimate agents from bots. Both networks are actively working with regulators and industry bodies to define interoperable standards for agentic commerce.
When will agentic payments be available to consumers?
Agentic payments are already available in limited form. All US Mastercard cardholders are expected to be enabled for Agent Pay by mid-2026, with global rollout following. Live agentic transactions have been completed in the US, Australia, New Zealand, Singapore, Malaysia, India, South Korea, and now Europe through the Santander pilot. However, widespread consumer adoption will depend on the maturation of AI agent platforms, merchant acceptance, regulatory clarity, and consumer willingness to delegate purchasing authority to AI systems. Most industry observers expect agentic commerce to become a meaningful share of transaction volume within three to five years.
MyValue Solutions is an independent publication. We are not affiliated with Santander, Mastercard, or any company mentioned in this article. This analysis represents our editorial assessment based on publicly available information and should not be construed as investment or financial advice.