EBAday 2025: Who's afraid of agentic AI?

Finextra

3 days ago

What potential does AI hold for the future of banking? What challenges and benefits do AI and GenAI-powered solutions pose? These questions were explored in an afternoon panel session at EBAday 2025.
0
This content has been selected, created and edited by the Finextra editorial team based upon its relevance and interest to our community.
Emmanuel Baviere, financial services advisor, Microsoft; Isabel Schmidt, EPO payment enablement platform, BNY; Oliver St Clair-Stannard, head of go-to-market, RedCompass Labs; Sébastien Racinais, head of innovation, global cash management, Crédit Agricole CIB; and Thomas Peeters, head of Western Europe, SWIFT, spoke on the panel session: 'AI and GenAI – What's in it for FIs?', moderated by Silvio Villa, senior manager, data science and AI, Be Shaping the Future.
Baviere kicked off the panel session with a few slides comparing analytical AI to generative AI (GenAI) to left-brain and right-brain activities respectively, with one having a more analytical and direct approach, and the other acting more creatively and generatively. Baviere stated that there is a wave of employee productivity in using AI tools, such as ChatGPT. He noted the introduction of AI agents, that hold potential to streamline and industrialise payments processes under the management of humans.
St Clair-Stannard said that AI agents are the next hurdle for banks, as they need to be mindful about AI agents making autonomous decisions on behalf of banks:
'See the use cases for AML fraud screening, anything that really touches on production data and on the live payment flow, we need to be very careful. If you're working with human in the loop or expert in the loop, there's always that ability to account for the decisions that are being made. As soon as you are making decisions autonomously with an agent, or once you start introducing an LLM into the ability to write and underwrite lending, any bias in the model is going to be amplified.'
Adding to St Clair-Stannard's point, Schmidt said that humans act differently when they know that an AI has made a decision as opposed to another human; the behaviours are impacted. Peeter posited that SWIFT is not yet ready for AI bots and agents to make decisions, however, the humans that make the decisions are sometimes influenced by AI information, which may include another form of bias. St Clair-Stannard concluded that that the bottom line is trust, and that over time as AI models evolve, agents will take over these decisions as they will never be wrong.
Schmidt stated that BNY has built an AI hub that integrated multiple systems and therefore had varied capabilities. She described that the hub and spoke model has a central team that develops the technology and 80% of employees have been AI trained in biases and ethical behaviours to use AI across the company.
Schmidt highlighted how AI applications are streamlining everyday processes for employees, who were seeing and predicting how these applications can be democratised and that commercial processes can progress at a faster pace more efficiently.
She added: 'In the end, it is important to remember that we are stewards of our own business, right? So the decision of how aggressively you move forward with developing AI has to go hand-in-hand with how much you invest in your second-line function and your control function. It's not just the people who apply it, but the people who actually manage to do this scenario, and risk appetite of the of your outbreak of respective institutions that makes the difference."
SWIFT's Peeter detailed that AI is integrated in three places across the organisation:
Using Copilot to help the staff, Detect anomalies and fraud, and In collaboration with financial institutions and industry groups on how to internationalise AI.
Racinais detailed Credit Agricole CIB's AI journey; considering more complex use-cases with real-time outputs, he sees three key areas where AI can definitively make a difference:
To better understanding their clients – AI can be used to personalise customer services, analyse transactional data of clients, and gain insights into their behavioural patterns. Operational efficiency – AI tools can optimise processing, automate tasks, enhance customer experience while reducing risk, streamline speed of execution, prioritise requests, and analyse sentiments to detect urgency or frustration. Developing new services – AI can be used in fraud detection and financial crime compliance checks.
Racinais summarised that there are 'plenty of use cases with today's technology. AI is developing so fast and so quickly that we could imagine plenty of other use cases open up in the near future. I do believe we are converging towards a very hybrid model, where we still have human interaction when it is it is required, more self-care functionalities for customers, and potentially more machine to machine interaction.'