Swift has conducted experiments using privacy-enhancing technologies (PETs) to allow financial institutions to securely share fraud intelligence across borders. In one scenario, PETs enabled participants to verify suspicious account information in real time, potentially accelerating the detection of complex international financial crime networks and preventing fraudulent transactions before they occur.

In another use case, PETs were combined with federated learning—an AI model that trains locally at each institution without sharing customer data—to identify anomalous transactions. Trained on synthetic data from ten million artificial transactions, the collaborative model proved twice as effective at detecting known frauds compared with models trained on a single institution’s dataset. Participating institutions included ANZ, BNY, and Intesa Sanpaolo, with technology support from Google Cloud.

Rachel Levi, head of AI at Swift, said the experiments could significantly reduce the billions lost annually to fraud, allowing suspicious activity to be stopped in minutes rather than hours or days. Following these successful trials, Swift plans to expand participation and begin a second phase using real transaction data to demonstrate real-world impact. The cooperative continues to explore AI’s role in cross-border payments, currently managing more than 50 use cases across proof-of-concept, pilot, and live operations, including its AI-enhanced Payments Controls Service for small and medium-sized institutions.