Abstract
The aim of this work is to study the growing role of artificial intelligence (AI) and Machine Learning (ML) in the fight against money laundering and the financing of terrorism (AML/CFT). In the first part, we introduce the various concepts related to AI and present the technological revolution it is bringing about in many economic sectors, and in particular for banking and financial institutions.
In the second part, we present the traditional methods of money laundering, including those using cryptoassets, and we review the recent scandals that have led to a tightening of AML/CFT regulations in France. We also analyse current AML/CFT methods and the solutions available, before carrying out a quantitative analysis, based on recent data, of money laundering trends.
Finally, in the third part, we examine how AI and ML improve the efficiency and adaptability of AML/CFT systems in the financial sector. We then identify the challenges and limitations of integrating them into anti-money laundering programmes, before highlighting their role in optimising risk management and regulatory compliance. Finally, we highlight the importance of human intervention in ensuring the accuracy, ethics and efficiency of AI-driven AML processes.
Extract
[...] Since there is no universal model, identifying fraud becomes complicated and is often confused with legal transactions. In addition, the evolution of financial crime patterns makes rules-based systems outdated and difficult to adapt to. Under these conditions, the authorities have to make a trade-off between efficiency, which implies fast systems but with the risk of missing fraudulent transactions, and security, which requires precise analyses but is costly and time-consuming. However, the use of AI could solve these problems, if real data is readily available and if technical constraints are overcome. [...]
[...] Their role is to mix the cryptoassets of several users in order to hide their origin and destination. Although they are often associated with illicit activities, their primary purpose is to enhance the confidentiality of transactions. Between January 2021 and May 2022, they received more than billion each month, with a record $1.5 billion in April 2022 alone (see chart 1 below). Chart 1. Monthly total value received by mixers : 2017-2024 Source : Chainalysis (2024) Among the crypto mixers with the strongest growth in 2023 and 2024 were WasabiWallet, JoinMarket and Tornado Cash. [...]
[...] The question explores the need for human intervention in AI and ML-driven AML/CFT processes to ensure accuracy of AML/CFT operations and compliance with ethical standards. It highlights the essential role of human-machine collaboration, where human experts validate alerts generated by automated systems, interpret complex cases and ensure regulatory and ethical compliance. We propose the following plan. In the first part, we present the general concept of AI, with its various definitions, its technological and disruptive impact in many sectors and in particular in the banking and financial sector which interests us in the context of this dissertation. [...]
[...] Open source AI makes it possible to avoid the creation of a monopoly on the AI market dominated by a few American giants. Open source AI offers several advantages to the EU in repositioning itself in the global AI race. Firstly, the code for the algorithms is accessible to everyone, enabling an international community of researchers, engineers and firms to study, develop and enrich the models. This process reduces the bias of the algorithms and facilitates their adoption. On the other hand, open source AI is universal and simplifies the transition to AI for all economic players. [...]
[...] These two pieces of legislation provide a framework for information gathering, first-level assessment of AI, in-depth checks and the investigation of penalty cases. The supervisory authorities have generic powers that allow them to conduct on-site investigations and access source codes, for example. The regulation on AI will have to be in phase with the sectoral rules. For the ACPR, the arrangements for supervision and risk management (governance, internal control, etc.) are well known, bearing in mind that certain requirements, for example in the area of cybersecurity, are already covered by existing texts. [...]
