AI and detection: Technical promises, marginal effects, and the absence of measurable effectiveness
The promises of artificial intelligence in the fight against money laundering are immense. Yet the gap between technological promises and observable effects remains considerable. Perhaps because the principal problem remains one of measurement, leading us at times to reinvent old concepts and to revive debates that have already been extensively explored rather than producing genuinely new knowledge.
Artificial intelligence has become the new frontier in anti-money laundering efforts. For some, it will enable criminals to launder money faster, more discreetly and on a larger scale. For others, it will finally provide financial institutions and financial intelligence units with the means to detect illicit flows effectively.
In their article “Review of artificial intelligence-based applications for money laundering detection » (2025), Seyedmohammad Ousavian and Shah J. Miah review nearly a decade of research involving machine learning, deep learning, network analysis and other advanced techniques. Their conclusion is one of caution:
- Publications are multiplying.
- Reported performances are high.
- Models are becoming increasingly sophisticated.
Yet significant difficulties remain: a lack of reliable data, the scarcity of confirmed cases, results that are difficult to compare across studies, and the absence of any broadly generalisable solution.
To make progress, it is necessary to distinguish between three separate questions.
The first concerns the use of new technologies by money launderers themselves: the production of false documents, the automation of transactions, and the potentially unlimited complexification of financial flows and networks. Yet the overwhelming majority of global money laundering activity continues to rely on long-established mechanisms: shell companies, real estate, international trade, intermediaries, money transfers and personal relationships.
The actual weight of new technologies – not limited to artificial intelligence – in global money laundering remains difficult to assess. The literature documents potential uses extensively but provides relatively little evidence allowing their real importance in observed practices to be measured.
The second question concerns detection. Certain financial institutions are investing heavily in artificial intelligence. The expectation is that these systems will identify anomalies more rapidly and reduce false-positive rates.
However, a study by Oztas et al. (2024) shows that many of the difficulties encountered stem from more fundamental issues: the scarcity of confirmed cases, limited access to data, heterogeneous datasets and the difficulty of comparing the performance of one model against another. The authors also highlight problems relating to the interpretation of results, the integration of tools into decision-making processes and the operational validation of models.
In other words, current limitations arise less from the power of the algorithms themselves than from the quality of the available data and the inherent difficulties involved in observing the phenomenon under investigation.
The third question is perhaps the most important: the overall effectiveness of anti-money laundering efforts. Ronald Pol (2017, 2018, 2020) has consistently published on this issue. For several decades, anti-money laundering systems have generated ever-increasing numbers of alerts, reports and indicators. Yet we remain unable to measure their impact on the actual reduction of money laundering.
We speculate about two opposing developments:
- Criminals will become more effective.
- Authorities will become more effective.
What are their respective gains in effectiveness? Almost any answer appears plausible.
Beyond the conclusions of these studies, a broader observation emerges.
Discussions between specialists in artificial intelligence and specialists in the economics of crime often reveal a degree of mutual unfamiliarity. Computer scientists sometimes approach questions that have already been extensively studied within the economics of crime, criminology or intelligence studies. Conversely, economists and criminologists often attribute near-miraculous qualities to AI, overlooking the fact that these models are ultimately trained using much the same data as before.
Each community tends to reinvent concepts already familiar to the other. Some innovations presented as disruptive breakthroughs appear instead to be rediscoveries. Others revive debates that have already been extensively examined, imposing the exhausting task of endlessly revisiting the same refutations.
This observation should not be interpreted as a criticism of augmented analysis. Rather, it serves as a reminder that the quality of any detection system depends as much on the understanding of the phenomenon under investigation as on the sophistication of the algorithms employed. This is precisely one of the challenges addressed by the European ENSEMBLE project. Beyond the development of AI-based tools, the project seeks to integrate human expertise, heterogeneous data analysis and a deeper understanding of criminal mechanisms to reconstruct criminal activity configurations that often remain invisible when viewed solely through the lens of available data. From this perspective, the challenge is not merely technological. It is also theoretical, methodological and operational.
Written by Paul Labic. Laboratory for theoretical and applied economics (BETA), CNRS UMR 7522. Associate researcher at the research lab of the French police academy (ENSP)
REFERENCES
Mousavian, S., & Miah, S. J. (2025). Review of artificial intelligence-based applications for money laundering detection. In Intelligent Systems with Applications (Vol. 27). Elsevier B.V. https://doi.org/10.1016/j.iswa.2025.200572
Oztas, B., Cetinkaya, D., Adedoyin, F., Budka, M., Aksu, G., & Dogan, H. (2024). Transaction monitoring in anti-money laundering: A qualitative analysis and points of view from industry. Future Generation Computer Systems, 159, 161–171. https://doi.org/10.1016/j.future.2024.05.027
Pol, R. F. (2017). Five money laundering myths for lawyers to avoid. Lawtalk, (910), 39–41.
Pol, R. F. (2018). Anti-money laundering effectiveness: assessing outcomes or ticking boxes? Journal of Money Laundering Control, 21(2), 215–230. https://doi.org/10.1108/JMLC-07-2017-0029
Pol, R. F. (2020). Anti-money laundering: The world’s least effective policy experiment? Together, we can fix it. Policy Design and Practice, 3(1), 23. https://doi.org/10.1080/25741292.2020.1725366
