Abstract : The rapid development of Artificial Intelligence (AI), particularly Large Language Models (LLMs), presents significant opportunities and challenges for financial supervisors globally. The Czech National Bank (CNB) is actively exploring the potential of AI to enhance the efficiency and effectiveness of its licensing and supervisory functions, especially concerning the evolving regulatory landscape including MiCA and DORA. This paper presents practical insights from the CNB's pilot testing, discussing a broad range of applications including market integrity monitoring (e.g., finfluencers, sanctions screening) and automated document analysis for licensing and compliance checks. It provides a particular deep dive into a project focused on automating the assessment of licensing documentation, which demonstrated the potential for a dramatic reduction in processing time from an average of 80 working days to single-digit days. The paper highlights crucial lessons from this process, including the necessity of a "human-in-the-loop" approach to mitigate AI inaccuracies and a shift towards structured inputs (e.g., JSON) to overcome the limitations of simple prompt engineering. The findings underscore that while AI offers substantial efficiency gains, its successful integration requires a strategic, process-oriented approach and cannot replace expert human judgment.

Keyword : Artificial Intelligence, AI, Financial Supervision, Licensing, Central Banking, MiCA, DORA, AI, Artificial Intelligence, Capital Markets, CASP (Crypto-Asset Service Providers), Central Banking, ChatGPT, DORA, Financial Supervision, Finfluencers, Human-in-the-Loop, Large Language Models (LLM), Licensing, MiCA, Process Automation, Retrieval-Augmented Generation (RAG), Risk Management, Sanctions Screening