We recently saw an interesting speech from the UK on how artificial intelligence (AI) could impact financial stability.

In France, the regulatory authorities have also been looking at AI.

Here are some snippets.

A while ago, the French Prudential Supervision and Resolution Authority (ACPR) issued an important discussion document on the governance of AI in finance. It identified four interdependent criteria, similar to the cardinal points of a compass, for evaluating AI algorithms and tools in finance: appropriate data management, performance, stability and explainability. The discussion document also considered certain workshops that had taken place including one on internal models in banking and insurance. Rather than studying internal models as a whole, the workshop focused on credit granting models; both are related insofar as scores produced by those models can also be used to build risk classes, from which risk-weighted assets are computed.

A number of important points arose from the workshop that banks and other financial institutions looking into this area may wish to keep in mind. These are considered in 5.3.4 of the discussion paper with perhaps the most important point being that whenever AI is used in the construction of internal models, an essential consideration in the validation process is how to define triggering events for a model revalidation.

Perhaps the bigger point in the discussion paper is its comments concerning governance. Incorporating AI into business processes in finance inevitably impacts their governance. The discussion paper set out a number of areas that businesses should particularly focus on including human/algorithm interactions, initial and continuous validation processes and audit (analytical and empirical).

About two years ago, in April 2022, the Banque de France Econ Notepad posted an interesting entry on the use of AI or machine learning techniques and how they could allow banks to develop new credit risk models.

But AI is not limited to the banking and insurance sectors regulated by the ACPR. As regards the financial markets and investment firms the Autorité des Marchés Financiers (AMF) is exploring the potential offered by natural language processing technologies in the analysis of documents prepared by listed companies.

Last summer, the First Deputy Governor of Banque de France, Denis Beau, addressed an audience on the lessons that regulators and supervisors have learnt from digital transformations and financial system turbulence. Among other things the First Deputy Governor argued that it is essential that the role played by new technologies and social networks in speeding up liquidity stresses be properly taken into account by regulations. At the same time, in the digital age, with easier access to information, and rumours, and the ability to conduct banking operations instantaneously, regulatory assumptions, in particular with regard to the flight of deposits also needs to be reviewed.

And more recently, on 6 May 2024, François Villeroy de Galhau, Governor of the Banque de France, gave a speech, Innovation by central banks: the sooner the better. Banque de France have already deployed AI in several different areas. For example, it uses machine learning to help detect fraud in transactions with the French Treasury that are managed by the Banque de France; “neural networks” help staff analyse the probabilities of default of non-financial corporations in the proprietary rating system; and natural language processing tools enhance analytical capabilities within the ACPR. Banque de France will step up this activity this year.

The upcoming Capital Requirements Regulation 3 and the Capital Requirements Directive 6 will bring major changes to the way EU banks manage risk. Among other things the legislation makes adjustments to the structure of exposure classes, and to the methods for calculating risk-weighted assets using the Standardised approach and Internal Ratings Based approach. Last August the European Banking Authority noted that in the context of credit risk, machine learning models might improve the predictive power and are not new to internal models used for credit approval processes.

And finally, from a wider European Union perspective, there is the new AI Act which will come into force later this year. The AI Act will impact the EU’s financial sector in a number of ways. For example, AI-based creditworthiness assessments by banks are considered high-risk AI use cases and will therefore have to comply with heightened requirements for such AI applications. The AI Act will also introduce new requirements for so-called general purpose AI systems, including large language models and generative AI applications.