Research project develops solution for the automated detection of illegal financial flows – including verifiable minimum requirements for the use of artificial intelligence in sensitive areas

In the fight against money laundering, established software-based detection methods work imprecisely and often raise false alarms. Consequently, investigating authorities are frequently overburdened because they must follow up on every suspicion. In the new research project MaLeFiz (Machine Learning for the Identification of Conspicuous Financial Transactions), researchers are now working on a solution that uses machine learning – an artificial intelligence technique – to improve the search for illegal money flows and make it more precise so that fewer false alarms are generated. In addition, the project partners are developing minimum requirements and control mechanisms for AI solutions used in the financial industry. Furthermore, the results of the AI are to be made traceable. Project partners are Deloitte GmbH, the Fraunhofer Institute for Secure Information Technology SIT, the Martin Luther University Halle-Wittenberg, the University of Leipzig as well as the Center for Technology and Society at TU Berlin. The MaLeFiz project is funded by the German Federal Ministry of Education and Research and will run for three years. Fraunhofer SIT is leading the project and is responsible for the development of the AI-based tool. For more information, visit www.sit.fraunhofer.de/en/malefiz.

The analysis of financial transactions is an important component in the fight against money laundering. However, current analysis methods identify too many suspicious cases, all of which must be verified individually by specifically trained analysts. The responsible authority, the Financial Intelligence Unit or FIU for short, received around 300,000 reports per year in the past two years alone and currently still has around 290,000 not finally processed alerts. Artificial intelligence methods promise better analysis options, which means fewer false positives. The research team is using machine learning methods to develop the necessary technical solution.

Transparent decisions made by artificial intelligence

For the results of such analyses to stand up in court, the IT solutions must meet certain requirements. For example, the decisions of an AI must be comprehensible: The AI should not be a "black box" and simply spit out a list of suspicious cases. It should make transparent the basis on which it flags a suspicious case. Accordingly, the team is researching legal and ethical issues. One goal of the project is to develop a catalog of minimum requirements for AI solutions in the financial industry that can be verified, for example, as part of an audit – a kind of TÜV for applications using artificial intelligence. In order to find out what the AI should look like in practice and to take users' needs into account as far as possible, the project partners are conducting interviews, workshops and tests with users. These results will then be incorporated into a demonstrator, which will also be tested in banks under real-life conditions. The demonstrator, the catalog of minimum requirements, and other project results will be made available to the public after completion of the project in September 2025.