You Cannot Automate Away Trust: NatWest on AI Agents and Payments

Quick Summary

When an AI agent initiates a payment instead of a person, what has to change underneath? Simon Eacott, Head of Payments at NatWest Group, says the industry is still grappling with it. Guardrails, regulation and liability all need to be right, along with industry standards and interoperability. Agents moving money need the same rules, boundaries and accountability as people, with human oversight retained, because you cannot automate away trust.

What does the industry still need to get right?

Eacott says this is one of the things the industry is grappling with right now, and because it moves so fast, getting the foundations right is massively important. He lists:

• Guardrails
• Regulation
• Where liability sits
• Industry standards and interoperability

All of these, he says, are going to be key to reaching the right level of trust.

Should AI agents follow the same rules as people?

Yes. If AI agents are going to be moving money, Eacott says, they need the same rules, the same boundaries and the same accountability. He also wants to keep an element of human oversight, both at the customer end and inside the bank.

His key line: you cannot automate away that trust element, and that is exactly where banks come in.

What do customers expect when an agent gets it wrong?

Customers do not care whether they are dealing with a human or AI, Eacott says. If a mistake is made, they will expect the bank to put it right.

So banks have to get the technology right and do everything they already do with every payment today, just in a new paradigm where the agent is the interface into the customer.

FF NEWS TAKE:

This is the clip that lines up with the agentic commerce principles NatWest and five other banks published this week. Eacott's framing is the practical one: nothing about an agent changes who the customer blames when something goes wrong. That makes liability, standards and oversight the real product, and it explains why banks want to write those rules together rather than inherit them from whoever builds the agents.