Belgium's SWIFT trials privacy-preserving AI with 13 banks to double cross-border fraud detection in real-time tests

SWIFT published results from experiments with 13 international banks showing that artificial intelligence combined with privacy-enhancing technologies can improve real-time detection of fraud in cross-border payments while maintaining end-to-end privacy and security. The trials tested two approaches to secure collaboration. One use case used privacy-enhancing technologies to let participants verify intelligence on suspicious accounts in real time, aiming to accelerate identification of complex international financial crime networks and prevent fraudulent transactions from being executed. A separate use case combined privacy-enhancing technologies with federated learning, training an anomaly-detection model locally at each institution without sharing customer information. Using synthetic data from ten million artificial transactions, the collaborative model was twice as effective at identifying known frauds as a model trained on a single institution’s dataset. SWIFT intends to expand participation and then launch a second phase of tests using real transaction data to demonstrate the technologies’ impact on real-world fraud.