Artificial intelligence represents a fundamental shift in data interaction and decision-making for both private and public organisations. Operating across financial markets for over more than 30 years in Asia, the US, and Europe, the technology's rapid evolution is underscored by how terminology has shifted from decentralised finance and NFTs to large language models in the span of five years. The financial sector is exposed to artificial intelligence more than most other industries due to three core pillars: data, speed, and trust.
Finance runs on data, including client video identifications, building plans, and mortgage sales documentation used for lending and onboarding decisions. Much of this information consists of unstructured text, where modern artificial intelligence excels compared to conventional software by recognising patterns, processing new data types, understanding text, and generating recommendations at unprecedented speeds and breadth. This capability compresses time, allowing credit decisions that previously required days to be completed in seconds, which simultaneously narrows the window for catching errors. However, efficiency gains remain worthless if client, investor, and public trust is eroded.
Switzerland sits at the forefront of these developments owing to the diversity of its financial centre, spanning global players, regional banks, and fintechs that serve as a natural laboratory. As an integrated supervisor, FINMA tracks adoption across banks, insurers, and asset managers, having conducted industry surveys starting in 2021 and most recently in 2025 across roughly 400 institutions. Three key findings emerged from the latest survey. Momentum remains high, with institutions running approximately five artificial intelligence applications on average and developing nine additional ones, with larger firms operating at a higher intensity.
Adoption currently remains broad rather than deep, prioritising early-phase and less critical use cases such as client communication chatbots, text generation, and process optimisation. Over 90 percent of institutions using artificial intelligence rely on generative methods because they are quick to implement, easy to delimit, and manageable in impact. Core processes involving credit decisions and regulatory metrics are approached much more cautiously. Only about half of the participating institutions have embedded artificial intelligence into an explicit strategy, and few develop applications in-house, relying instead on a small group of external technology providers. This dynamic places stringent demands on governance and risk management, particularly if the sector grows dependent on identical models, creating stability risks from shared outages or flawed models.
Accountability cannot be delegated to an algorithm, particularly with the rise of agentic AI systems that plan, execute, and chain tasks independently. Organizations must retain clear responsibility for the purpose, limits, and outcomes of deployed applications. Furthermore, a significant portion of artificial intelligence risk stems from malicious actors utilizing the technology, positioning it as a dual-use tool. Advanced models can uncover software vulnerabilities and accelerate attacks, while Interpol data indicates artificial intelligence-enabled fraud operations are approximately four and a half times more profitable than traditional methods.
