AI in banking could create a governance gap, warns researcher Andrew Ayres

With a background spanning enterprise technology and banking research, his work focuses on how emerging technologies influence risk, accountability and decision-making in modern organisations.

As an AI speaker, Ayres examines how intelligent systems are changing the relationship between technology, leadership and corporate governance, particularly in highly regulated sectors such as investment banking and financial services.

Through his research into AI governance and financial systems, Ayres explores how organisations can integrate advanced technologies responsibly while maintaining strong oversight, transparency and accountability in decision-making processes.

In this exclusive interview with the SpeakOut Speakers Agency, Andrew Ayres discusses how his work blends academic research with real-world financial services experience, why AI governance has become a growing concern for banks and regulators, and how concepts such as Principal–Multiagent Theory help explain the evolving relationship between humans, machines and corporate decision-making.

Question 1: Your career sits at the intersection of enterprise technology, academic research and financial services. How did those experiences shape your work on AI and banking governance?

“Andrew Ayres: I think, like a lot of people, I came to it somewhat accidentally. I started my IT career in the 2000s, and one of the most important experiences I had was with the global research house Gartner. During that time in the Middle East, working with IT leaders, I decided to embark on an MBA and got a taste for the ways that academia and real-life experience merged.”

“Following that, I also began to reflect on emerging technologies such as AI and major events that had shaped the past few years, like the global financial crash. The impetus for my PhD was really: what if the investment bankers who drove the financial crash became responsible for training AI?

“Is there a risk that some of their biases, their leanings, their appetites for risk and greed could become baked into algorithms or machines in the future with unknowable consequences? And so that's how I've become, I guess, over 10 or 15 years, a bit of a boundary straddler in terms of academic and professional life.”

Question 2: The global financial crisis reshaped financial regulation worldwide. What led you to examine AI governance specifically within European investment banks?

Andrew Ayres: “I think what was clear to me was that the global financial crash changed things. It was a huge event that sent a bullwhip effect across the entire globe, not just on Wall Street but much further afield.

“In the following years, we heard new regulatory policy and government policy come out around the fact that banks would have to be managed in a different way and that the CEOs of banks would become vicariously liable for everything that happened in the layers below them.

“Now, when you think about AI, it's extremely complex. It's very rare that you get a CEO who is aware of bank policy governance but can also dive into the technical detail of AI, which is a very complex software and technology-related issue, yet they are liable for what happens.

“So my sense was there was going to be a governance gap here and that the policies created internally and the work external regulators were doing were essentially going to get watered down as they moved down through the enterprise, to the point where the people writing the software and creating the AI applications were completely divorced from the governance policies and the regulation.

“It wasn't a guiding principle for them, and that creates the gap I talked about, the drift between what a bank CEO wants you to do and what you actually do on the ground.”

Question 2. Your work introduces the idea of Principal–Multiagent Theory. How does this concept extend traditional principal–agent theory in the age of AI?

Andrew Ayres: As a term it might sound a little nebulous, but I can explain.

“Principal–agent theory was a very famous economic theory from the 1970s. Two prominent scholars, Jensen and Meckling, came up with it, and essentially it's about the idea that a principal, the managing director or CEO of a company, decides on a set of activities that the agent, an employee, would go and do.

“It's formed the basis of corporate governance across many sectors, from banks to governments to healthcare providers.

“I believe that AI forms a new agent type. I don't believe it's just a software tool. I think that in time AI becomes a new agent type: a colleague, an employee, something that can think and act for itself.

“So principal–agent theory needed to be reassessed for its validity and ongoing adequacy in corporate governance settings. That's why I came up with principal–multiagent theory, because the CEO of a bank now presides over a workforce that is made up of humans and machines working together towards the shared aims of the bank.

“It's important from a risk perspective, it's important from a performance perspective, and it's vital from a governance perspective.”

Question 4: In trading and risk management environments, where decisions increasingly rely on algorithmic systems, how does AI challenge traditional corporate governance models?

Andrew Ayres: “AI frustrates corporate governance models in a number of ways. I talked earlier about the fact that bank leaders are responsible for what happens in the strata below them.

“The decisions being made at the interface with the software development tools that give rise to AI are choices around data and process steps that are taken if event A happens rather than event B.

“The people developing these software tools have a choice in what data and what learnings they choose to train the machine.

“Some data sets might have contained an anomaly in the November of a particular year. Data scientists and software developers can choose to select data from December onwards of that year, discounting the anomaly. However, the anomaly could well have been something telling, something that might happen again.

“So there are choices being made at that interface that impact the ways those tools get developed, the ways that they begin to churn through data and the ways that they begin to act.

“I always think of this as something that disappears from what I call our temporal understanding. Some of the ways the machine works contain impenetrable logic. It's not easy for a human to go in and say, “Ah, that's where it's gone wrong. That's the data set that it's contorted or misinterpreted.” It all becomes entangled.

“That, I think, is the largest governance gap I found: individual choice can deviate from what you might reasonably expect as the right way to do things. Not out of maliciousness or evil intent, but because humans do things differently. We all come with our own preferences and approaches.

“With AI, given its power and potency, we need to make sure that all those choices are right before we press the big green button.”

This exclusive interview with Andrew Ayres was conducted by Tabish Ali of the Motivational Speakers Agency.