Money laundering is a profound global problem. Nonetheless, there is little scientific literature on statistical and machine learning methods for anti-money laundering. Officials from the United Nations Office on Drugs and Crime estimate that money laundering amounts to 2.1-4% of the world economy. The numbers denote that the estimated amount of money laundered annually is almost 5% of the global GDP, or $800 billion. Unfortunately, money laundering makes it difficult for honest businesses to compete, as launderers are often able to offer products and services at a lower cost than the market value.   

The large amount of money in this crime and the social issues justify prioritizing anti-money laundering (AML) initiatives. Most current AML regulations and recommendations target cash transactions since these are the most common at the beginning of the money laundering process. This type of bank transaction, normally carried out in person, favours monitoring, unlike subsequent virtual transactions, whose objective is to hinder the tracking of the money’s trajectory at the various stages that make up the basic cycle of the money laundering process. 

According to McKinsey and Company, 2022; KPMG, 2018, the deployment of new technologies for anti-money laundering (AML) and countering the financing of terrorism (CFT) has been hailed as a “game changer” and an “up-and-coming revolution”. New technologies in this field are often called regulatory technology (RegTech), for they promise to improve AML compliance and enhance the delivery of regulatory requirements through AI. 

OECD principles for AI application in AML 

In 2019, the OECD developed its guiding principles for promoting AI as an innovative and trustworthy technology that respects human rights and democratic values. The OECD principles treat AI as a general-purpose technology, not in the specific AML/CFT context.  

 First, they state that AI should promote inclusive growth, sustainable development and well-being (OECD Principle 1), a general and uncontentious objective that offers little specific guidance. Secondly, they state that AI systems should be compatible with “the rule of law, human rights, democratic values and diversity”.

The key challenge in this context is the establishment of effective mechanisms for informed oversight. Informed regulation and supervision require developing expertise in new technologies, understanding the risks and being able to implement appropriate risk mitigation measures. Therefore, AML supervisors should be able to acquire such expertise to effectively supervise regulated entities’ implementation of safeguards in their AI-based systems.  

The OECD principles also urge for more “transparency and responsible disclosure around AI systems” to allow people to challenge outcomes which is important as decisions made by AI tools are not always intelligible to humans who may be affected. Financial institutions should not aim to eliminate the human element of AML; rather, they should seek to free up resources for higher-risk cases, supporting their analysts with AI-based tools without abdicating their responsibility. 

EU AI Act and its application to AML 

AI constitutes a core part of the EU’s digital single market, and there is a need for new harmonized rules on the development and use of AI-based products and services in this market. The proposed act is the first comprehensive legislative initiative for AI by a major regulator to ensure the transparency and explainability of processes and outcomes, as well as human oversight, cybersecurity, data protection and respect for privacy.  

The core provisions of the proposed EU AI Act define a risk-based approach to AI, in which AI-related risks are classified as follows: 

•  Unacceptable risks, such as social scoring, which will be prohibited under the AI Act; 
•  High risk, such as the use of AI in employee recruitment, medical devices, etc., which will be permitted subject to ex-ante third-party assessment of conformity with AI requirements, taking into consideration relevant sectorial legislation; 
•  Low-risk AI systems with specific transparency obligations, such as human impersonation by chatbots, will be permitted subject to the obligation to notify humans that they are interacting with an AI system; and 
•  Minimal- or no-risk AI applications will be permitted with no restrictions. 

In the AML context, the provisions of the proposed EU AI Act will also apply to the use of AI by regulated entities and FIUs. The act and its risk classification are compatible with the FATF San Jose and OECD principles, which advocate for positive and responsible AI innovation, fairness, transparency and accountability. The use of AI by regulated entities and FIUs in the AML context can be considered high-risk, as defined in Annex III sub-sections 6(e), (f) and (g) of the EU AI Act. 

Conclusion 

One hundred years ago, Nobel Peace Prize laureate Christian Lange noted that “technology is a useful servant but a dangerous master. In the era of AI, this is a justifiable concern not only for states but also for the world’s largest corporations, including Big Tech companies and major financial institutions. The financial and IT industries have already started revolutionizing AML. The use of AI in AML can help detect and prevent money laundering by analyzing vast amounts of financial data and identifying suspicious activity promptly and accurately. 

Fighting Money Laundering With Statistics and Machine Learning

Deploying artificial intelligence for anti-money laundering and asset recovery: the dawn of a new era

Persona