Ladies and Gentlemen
It is a pleasure to be here at the Point Zero Forum — an event built on exactly the policy and technology dialogue that finance now needs. SupTech — supervisory technology, the use of technology to modernise financial supervision — sits at that intersection.
We are in year four after the public launch of ChatGPT. Whatever one expected back then, the direction is now clear: artificial intelligence is a foundational shift — in how we interact with machines and data, and in how decisions are made.
Finance is particularly exposed to the rise of AI, not least because it runs on speed, data and trust. Advanced algorithms and machine learning have already quietly been transforming financial operations in the background for years. From a supervisory perspective, three areas warrant our attention.
The first is speed. AI compresses the timeframes our markets rely on. Price discovery that sometimes took minutes now occurs in milliseconds. Credit decisions once measured in days can now be settled in seconds. The efficiency is real. But so is the narrowing window in which an error can be caught before it spreads.
The second is market structure. The most capable models, and the computing capacity behind them, rest with a small number of providers. Suppose much of the industry comes to depend on the same few models, for risk scoring or fraud detection. Then a single outage, or a single flawed model, would no longer be one firm's problem, but potentially a matter of systemic stability.
The third is governance. AI places new demands on how institutions are run from within. Accountability is the bedrock of governance. It cannot be delegated to an algorithm and it requires that decisions are explainable.
Speed, structure, governance. We have to get all three right, because everything depends on the trust on which the financial system runs.
Switzerland and Zurich as an AI hub; FINMA surveys
At the same time, AI is not new to finance. FinTechs — but also banks and insurers — have used it successfully for years. That is especially visible here in Switzerland. Beyond finance, the country — and Zurich in particular — has become a remarkably concentrated hub for artificial intelligence, data science and robotics. Talent from around the world is drawn here by universities that rank among the global leaders in AI and this has made the region a magnet for global industry leaders — and it is one reason Switzerland will host the World AI Summit in 2027.
As an integrated supervisor that brings banks, insurers and asset managers under one roof, FINMA has a front-row seat to how this is transforming finance. Even before ChatGPT was released, we conducted the first of what are now three surveys on AI in the Swiss financial industry. Two findings stand out. The industry is highly dynamic: for every use case already deployed, two more are in development. And adoption is becoming more holistic — use cases no longer sit in isolation in the back, middle or front office, but increasingly extend across the entire value chain.
Dual use: the industrialisation of fraud
But new capabilities also bring new risks such as cyber risk or third-party dependencies. The very efficiency gains AI delivers can be turned to harmful ends. Fraud is the clearest illustration. According to Interpol's 2026 Global Financial Fraud Threat Assessment, AI-enhanced fraud is already four-and-a-half times more profitable than traditional methods, with agentic systems now able to plan and run entire campaigns on their own. Interpol calls it the industrialisation of fraud. We can only design measures against such threats once we understand them; and we can only respond in time if we have the right tools and analytical capabilities at hand. This is a core use case for SupTech.
The IOSCO SupTech survey
And here lies the deeper shift. One might say the real revolution currently is not in finance, where AI has been at work for years, but in supervision. AI can transform the very way we process data, derive insights and reach decisions. That goes to the heart of how we do our work.
This is the central message of a survey published only last week by IOSCO, the International Organization of Securities Commissions, which brings together the securities regulators of more than 130 countries and covers over 95% of the world's securities markets. Over the past three years IOSCO has launched a strategic initiative in SupTech, which FINMA has had the honour to chair.
Broad use cases; efficiency and effectiveness
So what does SupTech actually mean in practice? Based on this global stocktake, covering roughly three-quarters of global markets, the use cases are broad, ranging from automated document analysis to market abuse investigations, sentiment analysis and the monitoring of crypto exposures. For all their variety, they fall into two groups, reflecting two main goals of SupTech: efficiency and effectiveness. Let me make each concrete with an example from FINMA.
Efficiency – as an example: AI-assisted desk reviews
Efficiency essentially means doing what supervisors have always done; but doing so faster, and at lower cost. Take our on-site inspections. Before each one, the volume of documents to read keeps growing. So we are developing a generative AI use case that flags passages worth a closer look in this ever-growing pile of documents, before the actual on-site visit even begins. It is trained on the patterns of a large number of past inspections, on what FINMA staff have, over time, learned to notice. The tool runs on our local infrastructure and is currently still at proof-of-concept stage.
What I find instructive is how it is built. One model proposes the anomalies; a second, independent model checks each suggestion: does it hold up against the text or is it just a hallucination? Only suggestions that survive that scrutiny reach the supervisor. And one point matters above all: the tool never makes a decision on its own and never acts. It reads, and it suggests. The judgment, and with it the responsibility, stays with our people. That is not a technical detail; it is a principle.
Effectiveness – as an example: market abuse
Effectiveness is the more fundamental shift. Two-thirds of authorities look to SupTech for deeper insight, for capabilities they did not have before. It is the move from automation — the same tasks, only faster — to what one might call cognitive augmentation: seeing what was previously invisible.
Take the detection and enforcement of insider trading. It can be painstaking work. First, you have to spot suspicious trading patterns in a trove of transaction data, in the window before a big price move. Then you have to show those trades were actually informed by inside information: say, by tracing how it flowed from primary to secondary insiders across a network of relationships. We are currently developing AI-powered tools that support both tasks, using supervised machine learning models and network analysis.
The use cases that stand out: digital assets – e.g. near real-time crypto monitoring
Beyond efficiency and effectiveness, there is one use case that stands out globally: digital assets. As the chart showed a moment ago, this is where the interest of authorities most clearly outpaces their current activity.
And the case keeps growing in importance. The token world is expanding in breadth as well. Alongside established crypto-assets like Bitcoin, stablecoins are multiplying: their estimated circulating supply now tops USD 300 billion, up tenfold in just six years. Meanwhile, we are also seeing accelerating tokenisation of securities and other real-world assets.
Precisely because interest runs ahead of activity, this is where working together pays off most: on a technical level, supervisors worldwide are united by the same challenges and opportunities. This is where the IOSCO SupTech Forum comes in, which I have the privilege of chairing. Beyond the global survey, the Forum also delivers in practice with a pilot SupTech Sprint. FINMA hosted it here in Zurich just yesterday, in a hybrid format.
We brought together more than 100 technology and policy specialists from around the world, representing roughly two-thirds of global securities markets by value. Together, we launched the first phase of building tools for crypto-market supervision. After this successful launch, the build phase will follow in September.
Today, allow me to show you one of FINMA's own data-based supervision projects. We have built a dashboard that combines the quantities reported each quarter with daily market prices — token by token, matched through the Digital Token Identifier, the crypto equivalent of the ISIN. The result is a near real-time view of exposures, for the system as a whole and for each institution. It serves as an early warning for two risks in particular: concentration, where too much may rest on a single institution, and operational risks of tokens on a single blockchain such as Ethereum or Bitcoin. The first version is already up and running; we are extending it with more features and plan to include all the licence types we supervise in the medium term.
These are only some of many examples, but I hope they make the abstract concrete: for us, SupTech is no longer a promise — it is already at work. And across the survey, the motivation is unambiguous. Efficiency and effectiveness are rated a high or critical priority by 91% of authorities; reducing headcount, by just 4%. The purpose of SupTech is to empower human supervisors, not to replace them.
Three findings across use cases and regions
Beyond the individual use cases, the global picture shows three patterns.
First, despite all the progress it is still early days. There are many use cases, but most are as of now only partly implemented. Strikingly, smaller authorities and those in growth and emerging markets often move fastest, treating SupTech as a chance to build supervisory capacity.
Second, SupTech has moved from experimentation to strategic priority. It has left the IT back office and now is driven directly by Chairs and CEOs. That matters: given its reach, AI should be treated as a strategic question, not a mere operational tool. The commitment is becoming financial, too. Just over half of jurisdictions now have a dedicated SupTech budget; more still among smaller authorities and emerging markets. The upfront investment can be substantial; but the cost of inaction would likely prove higher.
Third, when it comes to the toolset supervisors use, AI clearly stands out as the big enabler, ahead of better data access and cloud computing. Importantly, adoption is not a vendor-led technology push: only a minority say the mere availability of tools drives it. This is strategy pull: real supervisory needs shaping the technology, not the other way round. The flip side, of course, is risk: here, cyber and data security are clearly named as the authorities' main concerns, ahead of third-party dependencies.
Modernisation of supervision through SupTech
Let me close with this: While AI has long been used across the financial industry, its impact is now increasingly being felt in supervision itself. This is where SupTech holds a unique potential to modernise our work, and the benefits are already tangible. By automating labour-intensive processes, we can reduce administrative hurdles and gain agility and speed. That does not merely cut costs; it frees up resources for where they matter most. And with sharper analytical insight, we will make our oversight not just faster, but deeper.
Yet, the limiting factor for AI in supervision will not be the performance of the latest model. It will be our ability to deploy these tools strategically, to oversee them effectively, and to ensure they serve the human in the loop. This modernisation of supervision is, above all, in our hands as supervisors. It is an opportunity we must seize.
Thank you.
