Wandhöfer tells that AI agents break payment authentication

At Open Banking Expo UK and Europe 2026 in London, Wandhöfer gave the opening keynote on day 2. Wandhöfer told that AI agents will select purchases and payment methods and send payments. The customer will not be in a bank app. Wandhöfer tells that this changes who controls payments and where the value goes, but gave no data for it.

The safety problem

Wandhöfer tells that AI agents have no human biometrics, and they are different from human users. Strong customer authentication is the rule that the customer agrees to each payment, and it must change for the authority that a customer gives to an agent. Wandhöfer told of prompt injection, where bad instructions make an agent do a task that it was not made for.

The business problem

If an agent selects the product and sends the payment, the customer can stop to use the bank. Wandhöfer tells that this makes the advantage of the customer interface and the brand less strong. It is not clear who will control the layer that makes the decisions.

What is not known

Wandhöfer tells that the change will not be fast. Rules, divided data, systems that are in use and consumer protection are obstacles. Wandhöfer gave no dates.

Insight: Agent identity is becoming a race

On 18 October 2026, World made an ID tool for agents that move money, and gave no test results. HKT Payment will test an agent ID in Hong Kong, and the ID does not stop a hack that uses AI. Wandhöfer wants agent identity and consent rules that machines can read. This reads as a race to set the agent ID standard: the first ID that banks accept can set the rules.