Resilience, AI Agents and Fraud: Pay.UK on Payments in an AI-Driven Economy

Quick Summary

Asked what the Sibos 2026 theme, Digital Finance for AI-Driven Economies, means from where Pay.UK sits, David Crawford, Chief Strategy and Transformation Officer, sets out three areas: using AI to make payments even more resilient, preparing for agentic commerce, and facing the dual threat and opportunity AI brings to fraud and cyber.

How can AI make payments infrastructure more resilient?

Resilience comes first, because Pay.UK's priority is to keep payments flowing no matter what, not just within Pay.UK but across the entire ecosystem. Crawford wants AI to help it scan the environment every minute of the day for signals and threats, rather than learning about developments once a week, so incidents can be detected and prevented before they happen. If prevention fails, AI-driven scenario planning means playbooks can be run there and then. He also raises resilience by design: enriching the data carried in the payment message itself, for example recording whether a payment originated from an AI agent or a human.

Is the infrastructure ready for agentic commerce?

Crawford frames agentic commerce with a road analogy and two questions:

• Are the roads wide enough? Broadly yes, he says. Volumes are not surging yet, but spending could fragment as, for example, your fridge starts ordering things itself, creating many more payments.
• Do the rules of the road need to change? If an agent made a payment, how should it be authorised, and who is liable? If an agent sends someone's life savings to a scammer, is it the person who authorised the agent or the company running it? These are difficult questions the industry will have to answer.

Why is AI both a threat and an opportunity?

The third area is what Crawford expects to be a huge discussion at Sibos. AI is already helping to protect against fraud across banks, Pay.UK itself and consumers using apps that keep them safe, and the same applies to cyber. But the industry has to be alive to the fact that the same tools can be turned against it, and build its defences accordingly.

FF NEWS TAKE:

The liability question is the one to watch. Capacity is an engineering problem an operator can plan for; deciding who pays when an authorised agent is tricked is a policy problem that touches consumers, banks, AI providers and regulators at once. Pay.UK putting agent-originated payments and message data on the table now is sensible, because rules written after the volumes arrive tend to be written in a hurry.