The Monetary Authority of Singapore (MAS) is validating artificial intelligence (AI) models for detecting financial fraud using real bank account information and transaction data.

A fintech outlet, Finextra, reported on May 11 that MAS is running a joint project with Singapore’s Government Technology Agency and the Singapore Police to boost fraud detection capabilities.

The core of the project is proactive fraud detection using AI and machine learning. In an ongoing proof-of-value experiment, MAS is pooling historical transaction data from five banks and training and evaluating models using participating banks’ actual account numbers. It aims to determine whether fraud signals can be detected better by reviewing banking-sector data together rather than relying on each firm’s separate detection systems.

MAS said the proof-of-value is designed to review proactive fraud detection using AI and machine learning techniques. It also said it is assessing the potential to improve models by using the participating banks’ real account numbers. This means it will use real data from the experimental stage to examine both detection accuracy and on-the-ground applicability.

Given the sensitive scope of data use, it also put security safeguards in place. MAS provided banks with a secure data-sharing environment where policies and procedures to protect customer information are applied. It also built a separate secure data-sharing system with industry participants. Account numbers were hashed so that only the bank that provided the data can identify its own accounts.

This structure is seen as an attempt to reconcile data integration between banks with personal data protection. It maintains the linkages needed for model training and performance validation without exposing real account numbers as they are. As a result, the experiment also tests whether a shared data-use framework for the financial sector can operate, beyond a simple technology validation.

MAS said the proof-of-value adds and complements AI and machine learning to existing financial crime prevention efforts at each financial institution and serves as a foundation for deeper industry collaboration. It is closer to expanding the scope of detection through shared data and additional models than replacing fraud detection systems already in operation at individual institutions.

MAS also left open the possibility of expanding the scope. It said that after assessing the effectiveness of the experiment and reflecting the training results, it could expand the model’s scope and sophistication by including a broader dataset and more diverse use cases. This is intended to further strengthen the financial system’s defensive line against crime.

In this context, two points narrow the focus of the project. One is whether fraud detection performance actually improves when historical data from multiple banks are combined. The other is whether hashed processing and a controlled sharing environment alone can secure a protection system strong enough for the industry to sustain real-data collaboration. MAS is expected to decide, based on the proof-of-value results, whether to further widen the use of AI in responding to financial crime.

The validation focuses on whether fraud detection accuracy can be improved by jointly using transaction data that are scattered across banks. Using real account numbers while leaving identification authority only with the providing banks stands out as a structure that validates both data use and information protection.