LLMs in finance ILB ESMA Turing Report

Report Overview

Title: Leveraging LLMs in Finance: Pathways to Responsible Adoption
Published by: ESMA, the Institut Louis Bachelier, and The Alan Turing Institute
Date: May 2025
Context: Based on a closed-door workshop held in June 2024, gathering 40 experts from regulatory bodies, academia, and the financial industry.

Purpose of the Report

This report offers a structured overview of how Large Language Models (LLMs) are being explored, used, and questioned within the financial sector. It aims to:

• Highlight current and emerging use cases of LLMs in finance
• Identify risks and challenges associated with their deployment
• Outline key areas for trustworthy and responsible adoption

Key Themes Explored

1. Use cases: internal applications (reporting, compliance), customer-facing tools (chatbots, robo-advisors), market analysis, ESG reporting, and more
2. Risks: from hallucinations and inconsistent outputs to security, fairness, and explainability concerns
3. Technical challenges: lack of robustness, data dependency, model bias, and the absence of shared evaluation benchmarks
4. Governance: accountability frameworks, staff training, regulatory coordination, and sustainability (carbon footprint)
5. Regulatory landscape: the role of the AI Act, MiFID II, and other European and global frameworks

Conclusion

The report stresses the need for a shared understanding and common standards to ensure the responsible use of LLMs in finance. It calls for proactive collaboration between regulators, researchers, and market participants to align innovation with stability, transparency, and trust.

Supported by the research chair FaIR – Finance and Insurance Reloaded, led by Marie Brière, this report reflects the growing interdisciplinary efforts to bridge AI research and financial practice.

FaIR: Finance and insurance Reloaded : FaIR: Finance and insurance Reloaded