Krzysztof Dąbrowski, Managing Director of the Banking Supervision Division at the UKNF, spoke during the panel discussion ‘New experiences, new choices, new limits of confidence: payments in the world of algorithms and immersion’, organised as part of the ‘Future of Payments’ conference.

During the discussion, he pointed to the importance of security in the context of AI agents that may perform tasks, make decisions and initiate payment transactions on behalf of users. He has stressed that ‘security is one of the key drivers of user trust, and thus ensuring it can accelerate the uptake of agentic AI solutions’.

He identified three components that should ensure user security at the time of using such solutions:

• agent authentication – it is necessary to confirm the agentic identity and the agent’s unambiguous link to a specific user; 
• permission scoping – the user should be able to define the scope of agentic actions, including the type of products the agent may buy, the price and transaction limits, and the conditions under which the agent may operate without obtaining additional consent;
• permission enforcement – what is needed is a mechanism, independent of the agent itself, to verify whether the agent is acting within the limits of the agent’s actual permissions and whether such permissions are not changed or exceeded without authorisation. 

Krzysztof Dąbrowski stressed that what requires attention is not only payment authorisation, but also the control of agentic actions undertaken on the user’s behalf. ‘While transaction authorisation mechanisms are already known and widely used, the scoping of, and independent monitoring of compliance with, AI agents’ permissions represent areas that require further development,’ Krzysztof Dąbrowski said.

Photos by Polish Cashless Foundation