AI Agents Are Now Making Real Payments in Europe

The First AI Agent Payment in Europe Just Happened

On 2 March 2026, Banco Santander and Mastercard completed Europe's first live end-to-end payment executed entirely by an AI agent. Not a demo. Not a sandbox. A real transaction, processed through Santander's regulated payment infrastructure, with real money moving between real accounts.

For fintech developers and AI agent engineers building payment systems, this milestone changes the conversation. Agentic commerce — where AI agents autonomously initiate, authorise, and complete payments on behalf of users — has moved from concept to production reality.

And the scale is staggering: AI agents already influenced $262 billion in holiday sales in 2025. Visa predicts millions of consumers will use AI agents to complete purchases by the 2026 holiday season. Both Mastercard and Visa have live agentic payment pilots running across Europe, Latin America, and Asia Pacific.

How Mastercard Agent Pay Works

The technical architecture behind agentic payments is what makes this interesting for payment developers. Mastercard's Agent Pay, launched in April 2025, provides the infrastructure for AI agents to transact on payment networks. Here's the flow:

1. Agent Registration and Authentication

Before an AI agent can make a payment, it must be registered and authenticated on the Mastercard network. This is analogous to merchant onboarding — the agent gets a verified identity, a set of permitted operations, and transaction limits defined by the cardholder.

2. Consumer Consent and Limits

The cardholder sets predefined limits: maximum transaction amount, permitted merchant categories, spending caps per day or month. The agent can only operate within these boundaries. Think of it as a programmable spending policy — the kind of system that payment developers and crypto developers building smart contract wallets have been designing for years.

3. Transaction Execution

The AI agent — in Santander's case, built with Microsoft Azure OpenAI Service and Copilot Studio, orchestrated by PayOS — identifies a purchase need, selects the merchant, and initiates payment through Mastercard's network. The transaction flows through existing card rails, meaning the entire settlement infrastructure already works.

4. Verifiable Intent

This is the genuinely innovative piece. Mastercard and Google jointly introduced Verifiable Intent — an open-source framework that creates a cryptographic audit trail proving what the consumer authorised and whether the agent followed instructions exactly.

The specification builds on established standards from FIDO Alliance, EMVCo, IETF, and W3C. It employs Selective Disclosure, sharing only the minimum necessary information with each transaction party. The framework interoperates with Google's Agent Payments Protocol (AP2) and the Universal Commerce Protocol (UCP).

For payment developers, this solves the fundamental trust problem: if an AI agent makes a payment and something goes wrong, how do you prove what was authorised? Verifiable Intent creates a tamper-resistant record that all parties — the consumer, the bank, the merchant, and any dispute resolution process — can consult.

What This Means for Payment Developers in the UK

As a fintech developer building payment infrastructure at Radom, I see agentic commerce as the next major infrastructure challenge for our industry. Here's what payment engineers need to think about:

New API Patterns

Traditional payment APIs are request-response: a user clicks "Pay", your backend processes the transaction. Agentic payments are event-driven and asynchronous: an AI agent decides when to pay, within constraints, and your system needs to handle transactions that weren't directly initiated by a human in real-time.

This means payment developers need to build:

• Policy enforcement engines that validate every agent-initiated transaction against the consumer's predefined rules in real-time
• Asynchronous webhook pipelines that process agent payment events with idempotency controls and exactly-once semantics
• Audit logging that captures the full decision chain from agent intent through transaction settlement

The Consent Infrastructure Gap

Current Open Banking consent flows are designed for humans: redirect to the bank, authenticate with biometrics, approve the payment. AI agents can't do biometric authentication. The industry needs new consent models — and payment developers who can build them.

Variable Recurring Payments (VRPs) provide a partial solution: the consumer pre-authorises payments within agreed limits, and the agent operates within those limits. But the gap between VRP capabilities and full agentic commerce is significant. Developers who can bridge this gap — building consent management systems that work for both human and AI-initiated payments — will be in extraordinary demand.

Crypto Payments and Agent Commerce

There's a natural convergence between crypto payment infrastructure and agentic commerce. Smart contract wallets already implement the concept of programmable spending policies: allowlists, spending limits, time-locked permissions. This is exactly what agentic commerce needs on fiat rails.

Crypto developers who understand programmable authorisation are well-positioned to architect the consent and policy layers that agentic payment systems require. The skills transfer directly — you're building the same patterns, just on different rails.

Building Agentic Payment Systems With Rust

The reliability requirements for agentic payment systems are even higher than traditional payment processing. When a human initiates a payment, they're watching for the confirmation. When an AI agent initiates a payment, failures might not be noticed until reconciliation — which could be hours or days later.

This is where Rust's compile-time safety guarantees become critical:

Policy validation must be bulletproof. A bug in your policy enforcement engine could allow an agent to exceed its authorised limits. Rust's type system lets you encode spending policies as types, making it impossible to process an unvalidated transaction. Concurrent agent requests need safe handling. Multiple AI agents making payments simultaneously through your infrastructure means high-concurrency payment processing. Rust's ownership model prevents the data races that could lead to double-spending or missed transactions. Cryptographic verification needs performance. Validating Verifiable Intent signatures on every transaction requires efficient cryptographic operations. Rust's zero-overhead abstractions give you the performance needed without sacrificing safety.

The $262 Billion Opportunity

AI-influenced commerce is already a quarter-trillion-dollar market, and we're only at the beginning. The infrastructure being built now — Agent Pay, Verifiable Intent, agent authentication protocols — will underpin a new category of commerce where AI agents handle routine purchasing, bill payments, and financial operations autonomously.

For developers in the UK fintech ecosystem, the opportunity breaks down into several areas:

Agent payment gateways — middleware that connects AI agent frameworks (LangChain, Claude MCP, Microsoft Copilot) to payment networks. This is new infrastructure that doesn't exist yet at scale. Compliance and monitoring — AI agents making payments create new regulatory challenges. Transaction monitoring systems need to handle agent-initiated payments differently from human-initiated ones. KYC/AML screening needs to verify both the agent's identity and the authorising consumer. Dispute resolution infrastructure — when an AI agent makes a payment the consumer didn't intend, who's liable? The technical systems for recording intent, verifying authorisation, and resolving disputes are all being built from scratch.

Key Takeaways for Fintech Engineers

Agentic payments are production-ready. Santander and Mastercard proved it works in a regulated banking framework. Visa has completed hundreds of secure agent-initiated transactions. This is no longer speculative. The trust layer is the hard problem. Verifiable Intent and similar frameworks are solving how to cryptographically prove what a consumer authorised. Payment developers who understand both cryptographic protocols and payment infrastructure will be building this layer. Rust developers are uniquely positioned. The combination of safety-critical policy enforcement, high-concurrency transaction processing, and cryptographic verification maps perfectly to Rust's strengths. AI agent developers who understand payments will be in exceptional demand. The intersection of agentic AI engineering and payment infrastructure knowledge is a small but rapidly growing talent pool. If you can build both the agent and the payment rails it transacts on, you're exactly who the industry needs.

The age of AI agents that can spend money is here. The infrastructure to make it safe, reliable, and scalable is what fintech developers and payment engineers will be building for the next decade.