Artificial intelligence in banking and compliance

Where artificial intelligence is used

The first large-scale use of machine learning in Swiss banks was in compliance. The Anti-Money Laundering Ordinance of FINMA requires banks and securities firms to monitor their business relationships and transactions effectively and to operate a computer-based system for this purpose, which identifies transactions with increased risks according to criteria set by the institution.¹ Traditional monitoring systems work with fixed rules, such as thresholds for cash deposits or for transfers to high-risk countries. They produce large numbers of alerts, most of which turn out to be unfounded after review by a compliance officer. Machine learning models promise to reduce these false positives by learning from past cases which combinations of features indicate a real risk, and to detect patterns, such as networks of related accounts, that rules cannot capture. For a Lugano private bank with a client base concentrated in Italy, the patterns of interest include the round-tripping of funds through Italian and Swiss accounts, the use of domiciliary companies and the trade-based laundering described in Chapters G.I and G.III.

A second field is client identification and know-your-customer processes. Natural language processing systems read identity documents, extracts from commercial registers and corporate documents in several languages, check them for consistency and fill the client profile. Screening systems compare client and counterparty names against sanctions lists and lists of politically exposed persons, a task that is particularly demanding for Italian, Russian and Arabic names with variant transliterations. Adverse media screening uses language models to read news in many languages and to identify reports of investigations or convictions. For the sanctions regime described in Chapter G.II, which requires Swiss institutions to apply the lists published by SECO under the Embargo Act, automated screening is now standard practice.² A third field is the investment business: portfolio construction, the analysis of research, the preparation of suitability assessments and client reporting. The newest field is the general use of generative artificial intelligence by employees for drafting, summarizing and translating, which brings the technology into almost every function of a bank.

The FINMA survey of 2024 and 2025

FINMA surveyed about 400 supervised institutions between late November 2024 and mid-January 2025 on their use of artificial intelligence. Around 50 per cent of the institutions used artificial intelligence or had initial applications in development, and a further 25 per cent planned to introduce it within three years. The institutions that used artificial intelligence had on average about five applications in operation and nine in development. Ninety-one per cent of them also used generative artificial intelligence such as chatbots, which FINMA interpreted as a sign of heavy dependence on a small number of large technology providers. About half had embedded the use of artificial intelligence in an explicit strategy. The risks most frequently named by the institutions concerned data quality, data protection and explainability, followed by information security, accuracy and outsourcing. Larger institutions used considerably more applications than smaller ones, which typically relied exclusively on external providers.³

The last finding is of particular relevance to Lugano, where most banks are small or medium-sized and the independent asset managers and fiduciaries are small firms. Such institutions will rarely develop their own models. They buy compliance software, portfolio management systems and generative AI tools from providers, many of whom are located outside Switzerland, and the artificial intelligence is embedded in the product. For them the legal questions of artificial intelligence are largely questions of procurement, outsourcing and data protection.

Notes

1. FINMA Anti-Money Laundering Ordinance of 3 June 2015 (AMLO-FINMA), SR 955.033.0, Art. 20 para. 1 (effective monitoring of business relationships and transactions) and para. 2 ("Banken und Wertpapierhäuser betreiben für die Transaktionsüberwachung ein informatikgestütztes System"), with an exception in para. 4 for institutions with few relationships or transactions, text at https://www.droit-bilingue.ch/de-fr/9/95/955.033.0-20-24.html. ↩
2. Federal Act on the Implementation of International Sanctions of 22 March 2002 (Embargo Act, EmbA), SR 946.231, Art. 1 (compulsory measures restricting, among others, payment and capital transfers), text at https://www.droit-bilingue.ch/de-en/9/94/946.231-1-4.html; see Chapter G.II. ↩
3. FINMA, "FINMA survey: artificial intelligence gaining traction at Swiss financial institutions" (April 2025), https://www.finma.ch/en/news/2025/04/20250424-mm-umfrage-ki/ (24 April 2025). ↩

For general information only. This article does not constitute legal or tax advice, does not replace advice on an individual case and creates no lawyer-client relationship. It reflects the legal position as of August 2025.