The Swiss Financial Market Supervisory Authority (FINMA) has published remarks by Chair Marlene Amstad outlining Subsuma, an artificial intelligence (AI) pilot to support assessments of fintech firms’ licensing and other regulatory requirements. Prepared for the Swiss Bankers Association’s Digital Finance Day on Nov. 9, 2026, the remarks describe an extension beyond internal AI analysis toward assistance with responses to firms, while retaining human responsibility for decisions. The prototype uses three open source language models running on FINMA’s own hardware, including Switzerland’s Apertus. It summarises business models and assesses cases against statutory provisions that could trigger licensing requirements. Every statement must be supported by references to the case file. FINMA receives close to 100 fintech enquiries annually, and the project aims to accelerate responses. Initial testing found that larger models were needed for high quality output, with more powerful versions planned for production use. Every outgoing letter will require an expert’s review and signature. Amstad also emphasised that financial institutions remain accountable for AI applications, including those supplied externally. FINMA’s 2025 survey of around 400 institutions found that only about half of those using AI had an explicit AI strategy. Most relied on external providers, increasingly concentrated among a few large technology groups, creating potential financial stability risks from shared outages or flawed models. The remarks also called on firms to patch vulnerabilities faster and use AI to defend against attacks, highlighting gaps in digital fraud policies and assigned responsibility.
Swiss Financial Market Supervisory Authority outlines AI pilot to support fintech licensing assessments
In prepared remarks for the Swiss Bankers Association’s Digital Finance Day, Swiss Financial Market Supervisory Authority Chair Marlene Amstad outlined Subsuma, an AI pilot to assist fintech licensing assessments. The prototype uses three open source models on the authority’s hardware, with larger models planned and expert review and signature required for every outgoing letter. Amstad also stressed firms’ accountability for AI and risks arising from concentrated technology providers, cyberattacks and digital fraud.