SWIFT trials privacy-enhancing AI data collaboration with 13 banks to double cross-border fraud detection

SWIFT published results from AI-focused experiments with 13 international banks showing how privacy-enhancing technologies (PETs) could allow cross-border sharing of fraud insights without exposing customer data, improving fraud detection and response in international payments. In trials using ten million synthetic test transactions, one of the approaches tested was reported to be twice as effective at identifying known fraud compared with a model trained on a single institution’s data. The experiments tested two main use cases. In one, PETs enabled participants to verify intelligence on suspicious accounts in real time while maintaining end-to-end privacy and security. In another, participants combined PETs with federated learning, where an AI model trains locally at each institution, and the resulting model identified anomalous transactions with materially improved performance based on the synthetic dataset. Participants included ANZ, BNY and Intesa Sanpaolo, with technology partners including Google Cloud. SWIFT intends to broaden participation before starting a second phase of testing using real transaction data to assess the technologies’ impact on real-world fraud.