Resisting is not enough: central banks put to the test by artificial intelligence

By Sergio Boccadutri

This morning, at the tenth Annual Research Conference of the National Bank of Ukraine and the Narodowy Bank Polski in Kyiv, Fabio Panetta spoke about something that concerns all central banks, not only Ukraine’s. The Governor of the Bank of Italy opened his address with a concrete fact: the Ukrainian hryvnia turns thirty.

The national currency was introduced in 1996, at the end of a period of post-Soviet monetary instability, and since then it has weathered three decades of crises and, over the past four years, the test of a full-scale war. During the Russian invasion, with inflation climbing above 26 per cent and the official rate raised to 25 per cent, Ukraine’s banking system kept working and its institutional framework held. In February, the International Monetary Fund approved a new four-year programme that recognises this achievement. Meanwhile, the alignment of Ukrainian banking rules with European standards has continued to advance, rising from 50 per cent in 2022 to almost 80 per cent today.

It is a story of resistance and resilience, and Panetta uses it to recall what the word credibility usually means: holding firm. Defending independence, protecting the framework of monetary policy, preserving the capacity to act under pressure. Credibility, in this sense, is built over time and is measured when circumstances become extreme. Ukraine’s central bank is the perfect example of this virtue.

The challenge of adapting to AI

But Panetta’s speech serves to introduce a second question, of the opposite nature. Alongside the ability to withstand a shock, there is another, more uncomfortable one: the ability to understand a change and adapt to it. And here the testing ground is not war, but artificial intelligence.

The difference is substantial. Faced with a geopolitical crisis, a central bank knows what it has to do: stiffen its stance and defend the perimeter. Faced with the spread of AI, it cannot do the same. Economic models, supervisory tools and payment infrastructures were built for a different technological world.

Generative AI, Panetta notes, is spreading faster than any previous technology. In the United States, it reached half of all households in about three years, compared with the roughly ten years the internet needed and the seven years taken by the mobile phone. A transformation of this speed produces changes in productivity, work, markets and payments so quickly that those who have to govern them have little time to measure them.

AI creates a dilemma for monetary policy

The problem for a central bank is that AI pulls monetary policy in two opposite directions. In the short term, it pushes prices up: the investment needed to build data centres and train models drives demand, while at the same time raising the cost of what is needed to produce them, namely computing power, energy and specialised skills. As long as the supply of these factors does not keep pace, the pressure on prices remains.

In the long term, however, the dominant effect should be the opposite: a technology that makes labour more productive expands the economy’s productive capacity and, all else being equal, tends to bring inflation down. For Italy, the Bank of Italy’s estimates indicate that widespread adoption of AI could add more than one percentage point to productivity growth per year.

The conditional, though, carries weight: that gain depends on how deeply the technology is actually adopted, particularly in small and medium-sized enterprises.

Who benefits from AI?

Which of the two effects will prevail depends above all on labour, that is, on how the benefits of AI will be distributed. Panetta distinguishes two paths.

In the first, technology replaces people: it automates tasks that workers used to perform, reduces the demand for labour and shifts income towards capital. In the second, AI works alongside and supports workers: it creates new tasks, raises the demand for skills and lets people pocket part of the gains.

The consequences diverge. If the second path prevails, households feel richer and more secure, they consume more and the pressure on prices lasts longer. If automation prevails, uncertainty about jobs and wages pushes people to save and consume less, and the cooling effect arrives sooner.

Reality will perhaps sit somewhere in between, but with a complication: the gains are not distributed evenly. Larger firms adopt AI earlier and more effectively than small ones, and this risks concentrating markets even further.

For those who make monetary policy, the lesson is clear: it is not enough to estimate how much AI will raise productivity. One has to understand who those gains will go to, because their distribution determines demand and, therefore, inflation.

Europe’s adoption gap

There is then a second question, which concerns Europe: will AI widen the gap between countries?

Today, the United States is racing ahead. AI-related investment has become an engine of its growth and draws capital from the rest of the world: in 2025, foreign purchases of US equities came close to $742 billion. Europe, which still lags behind on AI development, has so far felt these effects to a much smaller degree.

But according to Panetta, the game is played not only on who builds the technology, but also on who adopts it. Even without becoming a major developer of models, the continent can use them just as quickly as other advanced economies.

The difference in the speed of adoption then becomes everything: the countries that integrate AI fastest secure a lasting productivity advantage, expand their market shares and attract investment, while the others watch their competitiveness decline.

It is the same knot Italy knows well: whether that extra point of productivity materialises or stays on paper depends almost entirely on how deeply the country adopts the technology, not on how fast the technology evolves.

The unknown natural interest rate

Panetta then ventures a cautious reflection that, for a central banker, is no trivial matter. No one today knows where the natural interest rate lies, that benchmark level which neither stimulates nor restrains the economy and which serves to guide decisions on rates.

Higher productivity would tend to raise it, but greater inequality, the precautionary propensity to save and the greater weight of capital in incomes could push it downwards.

In an economy that changes so fast, the Governor says, central banks should give more weight to what actually happens, such as demand and prices, than to estimates of variables that cannot be observed directly.

It is an invitation to trust facts more than models, and to the ears of those who build the models, it should sound almost like a change of course.

Markets are already pricing in the AI boom

What makes the picture more treacherous is the way markets are already pricing all this in. Share prices embody very optimistic expectations about AI’s future profits. Those valuations help technology companies raise capital and fuel new investment, but they expose markets to sharp corrections as soon as expectations are disappointed, as the recent volatility of technology stocks has already shown.

A transition that runs through this is unlikely to be smooth, and it is one more reason for the central bank to stay pragmatic rather than chase a number it cannot observe.

AI is also changing banking and payments

Artificial intelligence, moreover, does not test monetary policy alone. In banking supervision, it can improve credit assessment and fraud detection, but it also creates new vulnerabilities: when too many banks rely on the same models or the same technology providers, a single error or failure can spread and affect the entire system.

At the same time, in the payments sector, new players and faster services increase competition and efficiency, but they make the ecosystem more complex and fragmented, with novel operational and compliance risks.

The answer Panetta points to is not to reject innovation but to incorporate it into the existing architecture: a two-tier system in which central bank money remains the anchor and private money and payment services are the layer operating above it.

Adapt, do not dismantle.

Credibility means knowing when to change

And this is where the two questions raised in Kyiv come back together.

For decades, the credibility of a central bank meant one thing above all: holding a steady course on inflation, defending independence and not bending. The first question, then, is credibility, and Ukraine’s central bank has shown it in its harshest form, under war.

To this, Panetta adds another, of the opposite sign: the credibility that knows how to change, able to understand a transformation the central bank was not equipped to govern and to rework its own tools without losing sight of its mandate.

Resisting and adapting are not the opposite of one another, but two sides of the same challenge: staying credible when the world around us changes profoundly.