AI in finance: Opportunities, risks and the need for control

Good morning, dear friends and colleagues,

A very warm welcome to Epicenter, and to everyone following us online. It is a pleasure to open the third seminar on the Impact of AI on Economy, Finance and Supervision, organised jointly by the Bank of Finland and the Finnish Financial Supervisory Authority. Our cooperation with the Eurosystem Innovation Forum, Innov8, adds a further dimension. Thank you for joining us.

Let me begin with a story. You may remember The Sorcerer's Apprentice from Disney's Fantasia, originally inspired by Goethe's poem. A young apprentice discovers a wonderfully efficient way to carry water: he brings a broom to life. At first, everything works beautifully. The broom does all the hard work. But then comes the problem. The apprentice doesn't know how to stop it. And soon, the room is flooded.

I doubt our AI agents will start carrying buckets of water. But the story raises a remarkably modern question. Not long ago, we were asking what artificial intelligence could do for us. Today, we must also ask what we should allow AI to do on our behalf, and how we remain in control.

The lesson is not to reject innovation, but to match ambition with control.
Let me offer three perspectives: AI’s economic potential, its implications for financial stability, and why resilience and strategic choice matter.

AI's economic promise – and the productivity puzzle 

First, the economic opportunity. It is easy to be impressed by what the latest AI models can do. But from an economist’s perspective, a more important question is what these capabilities will ultimately do to productivity, growth and people’s working lives.

AI can already help us analyse huge datasets, find patterns, write and test code, and communicate complex results more effectively. In central banking, these are not marginal improvements. Better use of information can strengthen research, forecasting, financial supervision and, ultimately, the quality of our decisions.

There is also a very human benefit: if technology takes care of routine work, we can devote more attention to questions that require judgement, creativity and interaction with others.

But translating technological progress into productivity growth is not quite that simple. In 1987, Nobel laureate Robert Solow famously observed that computers were visible everywhere – except in the productivity statistics.

Almost forty years later, one might wonder whether we are about to experience a similar AI paradox.

And yet, history offers a reason for optimism as well as patience. The productivity gains from earlier general-purpose technologies eventually emerged, but only after businesses had adapted their processes, invested in new skills and learned how to make the most of the technology.

The question is not only what AI can do. It is what we learn to do differently because of AI.

The benefits will be uneven. Some tasks will disappear, others will emerge, and most will change. We cannot yet predict the net effect on employment. But we can help workers and businesses adapt.

I am optimistic about AI’s potential to raise prosperity. But optimism is not a forecast, and technological potential is not yet an economic outcome.

For central banks, this calls for intellectual humility. We should be ambitious in experimentation, but modest in claiming that we already know where this transformation will lead.

AI in finance: opportunities and systemic risks

Let me now turn to finance. Here, the opportunities are considerable – but so are the stakes.

The Finnish Financial Supervisory Authority published survey results just a couple of weeks ago. Of the financial-sector entities responding, 93 per cent already use AI or plan to do so within two years. The reported use of generative and general-purpose AI has risen from 74 to 90 per cent in a single year. AI agents are emerging as the next wave.

These figures tell us something important: this is no longer a discussion about whether AI will enter finance. It is already here – in internal processes and increasingly in services offered to customers.

Used well, AI can improve fraud detection, risk assessment, customer service and efficiency. It can also help supervisors identify vulnerabilities earlier and make better use of information.

But finance is different from many other industries in one crucial respect. It rests on trust. A mistake in a recommendation engine may be annoying. A mistake in a credit decision, a payment system or the management of liquidity can have very different consequences.

This means that governance cannot be an afterthought. Institutions must know where AI is being used, what it is allowed to do, how its performance is monitored and who is accountable when something goes wrong. Human oversight must be meaningful – not merely a signature at the end of an automated process.

For those of us concerned with financial stability, there is a further distinction. What is manageable for an individual institution may not be manageable for the system as a whole.

Consider a world in which many financial institutions rely on a small number of technology providers, similar AI models and overlapping data sources. Each institution may find its choices perfectly rational. Yet common dependencies create the risk of common failures. And similar models can make participants respond to the same signals at the same time.

Think of a bridge. It may safely carry hundreds of people walking at different paces. But if a much smaller group starts marching in step, their synchronised movements can put the bridge at risk.

The problem is not necessarily the weight. It is the synchronisation.

Financial markets can work in much the same way. What seems perfectly rational for each institution individually may become destabilising when everyone acts at the same time.

There is also a different type of concentration risk: the extraordinary growth of investment and financing linked to AI. Much of the valuation of leading technology companies rests on expectations of future earnings. And expectations can change quickly. On 29 October, Alphabet – Google’s parent company – lost close to half a trillion US dollars in market value in a single trading day, amid doubts about the returns on its massive AI investments. That is not, in itself, a systemic event. But it illustrates the scale of repricing that can follow a shift in sentiment. When exposures are large and interconnected, the consequences may extend well beyond the technology sector.

Cybersecurity adds another layer. In July, the ESRB issued a warning about systemic cyber risks stemming from frontier AI models. AI can help defenders identify weaknesses and respond to threats. But it can also make attacks faster, more sophisticated and easier to scale. We cannot simply assume that defensive capabilities will always keep pace.

None of this is an argument for slowing innovation to a standstill. It is an argument for bringing risk management into innovation from the start. An AI application is not truly successful if it is efficient in normal times but cannot be understood, controlled or safely replaced when conditions change.

Digital sovereignty is about preserving choice

This brings me to my third perspective: resilience and digital sovereignty.

Having worked with payments and financial market infrastructure for many years, I have come to appreciate a simple truth: when everything works, resilience hardly gets noticed. But when something goes wrong, it suddenly becomes everyone's priority. That lesson applies just as well to AI.

Europe benefits from an open global technology market. But openness should not leave us without choices. We must know where critical functions depend on a single provider or infrastructure beyond our control – and whether we have credible alternatives.

Digital sovereignty, as I see it, is not about doing everything ourselves, or isolating Europe from global innovation. It is about retaining the capacity to make our own choices. That requires expertise, access to technology, the ability to switch providers and, for the most critical services, credible fallback arrangements.

And it requires proportionality. The same safeguards are not necessary for a tool that helps draft a document and for a system that supports critical financial infrastructure. What matters is the consequence of failure, the degree of concentration and our ability to continue operating.

Two practical examples come from very different fields. Last month, Nokia and ICEYE announced plans to develop secure satellite communications based on European technology, giving governments greater control when terrestrial networks are disrupted.

The ECB, meanwhile, has announced that the Eurosystem’s Integrated Reporting Framework will be implemented using a European cloud. Both initiatives illustrate how strategic autonomy can be translated into practical technology choices.
Yet supply alone is not enough. For credible European alternatives in AI and cloud services, providers must be able to scale. That also requires demand, including through cooperation, procurement and the willingness to use competitive European solutions where they meet the need.

Payments provide a useful parallel. In Finland, the efficiency of international card networks has served consumers and businesses very well. At the same time, our reliance on them makes it essential to think about continuity and alternatives. The proposed digital euro is one way the Eurosystem aims to strengthen Europe’s choice and resilience in retail payments, alongside cash and private-sector solutions.

The broader lesson for AI is the same: alternatives are much easier to build before a disruption than during one.

Friends and colleagues,

Over the next two days, our excellent speakers will help us explore what AI already delivers, what it promises, and what remains uncertain. The programme ranges from global AI markets and financial stability to central-bank applications, with perspectives from across Europe and the Bank of Korea.

I hope we will keep three questions in mind. Where can AI genuinely improve outcomes, rather than simply accelerate existing processes? How can we capture those gains without creating new system-wide vulnerabilities? And how do we preserve human accountability and strategic choice as technology becomes more capable and more autonomous?

There is no reason to choose between innovation and trust. In finance, trust is what makes sustainable innovation possible.

And, to return to The Sorcerer’s Apprentice, the lesson is not to leave the broom in the cupboard. It is to make sure we know how to stop it. We should welcome technologies that can change how we work, be demanding about how they are governed, and build resilience before it is put to the test.

Thank you to our speakers, to our partners and to everyone who has helped organise this event. I encourage all of you – here at Epicenter and online – to contribute actively to the discussion.

I wish you an inspiring and productive seminar.