ESMA-Turing-FaIR Workshop: An Exploration of LLMs in the Financial and Insurance Sector

Large Language Models (LLMs) appeared on the public scene at the end of 2022, notably with ChatGPT, developed by the American company OpenAI. Since then, the range of offerings has continued to expand, with general solutions proposed by major tech companies, specialized models, and internal solutions developed by large groups in various sectors.

Finance and insurance sectors are not exempt from the crucial question: “What are the opportunities and risks associated with LLMs?” In an attempt to provide an answer, the FaIR program of the Institut Louis Bachelier, together with ESMA and the Turing Institute, organized a half-day closed door workshop on June 20th. This workshop brought together top experts in the field: researchers, regulators, and data and tech managers from major financial institutions. The objective was to examine the opportunities related to the use of LLMs in the financial and insurance sectors while highlighting the main risks and challenges associated with their use.

The discussions focused on evaluating the robustness, security, fairness, and integrity of LLMs. As in many other sectors, LLMs can bring significant value to finance and insurance. However, in such a crucial and regulated sector, it is essential to measure and consider the risks generated by the use of these models to balance the benefits gained. Representatives from the financial industry presented an overview of their external and internal LLM policies. Regulatory bodies demonstrated an advanced understanding of these models and the associated risks, as reflected in the AI Act, a European regulation adopted in May 2024.

A report summarizing the workshop discussions will be prepared in the coming months, aiming to reflect the sector’s overall stance on LLMs and the various approaches that may be taken.