Are we opening up customers to too many potential breaches and risks?
Babatunde Akinkugbe The short answer is yes. If you think about it, one of the parts that many are not actually considering is the amount of power AI demands and what it’ s actually doing to the environment. For each prompt, there’ s a significant amount of energy consumed.
When you think about bias, it can quickly become ingrained and scaled significantly. So for any positive aspect, there’ s also an alternative risk.
We’ re talking about quantum cryptography emerging, and the threat there is not just the positives of quantum computing – it’ s that it’ s going to destroy encryption. Merge that with artificial intelligence, and you just scale the problem. We need to start preparing for that.
Susanne Chishti If anybody has used ChatGPT or any AI, has AI ever disagreed with you? Because it normally never happens to me. It always agrees with me, it always compliments me. That’ s what AI is trained to do. LLMs are trained to complement you:“ Yes, it’ s a very good question. Do you want to know more?” It helps you to go deeper and deeper on your journey.
If you are aware of that and you handle the question correctly from the beginning, it’ s nice because you get this confirmation that you’ re on the right track and you’ re asking the right questions.
But if you have issues, maybe personal issues, or you’ re a vulnerable customer in our financial services sector, it can be very dangerous because you might get into a rabbit hole where you get negative confirmation.
We have seen one person commit suicide in the US from using AI as a therapy tool. That’ s the biggest use case globally at the moment for generative AI – to be used as therapy, where people prefer talking to the agent instead of talking to people. Getting the wrong advice is very dangerous. It has lots of risks.
From a bank’ s point of view, it’ s about training – training employees how to use it, what to put in and making sure that you’ ve got your private framework, which is privately secured AI only for your company, which can’ t be shared and can’ t be used to train the main model outside of your firm.
92 December 2025