Global financial messaging coordinator Swift has launched two AI pilot projects to target cross-border payments fraud, which it says cost the financial industry US$485bn in 2023.

The first project brings together 10 major financial institutions, including BNY Mellon, Deutsche Bank, DNB, HSBC, Intesa Sanpaolo and Standard Bank, to see if AI can be used to identify potential fraud patterns from anonymously shared datasets.

Swift hopes that using a model known as federated learning, in which an AI model is trained on multiple locally held datasets, will facilitate confidential information sharing among financial institutions.

A spokesperson for Swift tells GTR the pilot aims to “not only test the efficacy and business value of agreed use cases to improve fraud detection rates and reduce false positives, but also thoroughly test technologies and processes such as secure data collaboration, privacy-preserving controls and federated learning to ensure all data remains anonymous, secure and cannot be re-engineered out of any federated trained AI models”.

For the second pilot, Swift aims to train an AI model to improve its payment controls service, which screens financial institutions’ outgoing payment messages to detect potentially fraudulent activity.

The model will use historical Swift data, which will be further refined by customers following tests on their live traffic data.

These experiments are expected to be ongoing through much of the rest of the year. If successful, the pilots should remain usable in the future, with Swift “putting in place mechanisms that will ensure the adequacy of models used over time”, including drift monitoring that will “detect if a previously well-fitted model still produces the expected results”, it says.

The two projects are also aligned with emerging global AI standards such as ISO 42001, the NIST AI Risk Management Framework and the EU AI Act.

Tom Zschach, chief innovation officer at Swift, says of the projects: “AI has great potential to significantly reduce fraud in the financial industry. That’s an incredibly exciting prospect, but one that will require strong collaboration.

“Swift has a unique ability to bring financial organisations together to harness the benefits of AI in the interests of the industry, and we’re excited by the potential of both of these pilots to help further strengthen the cross-border payments ecosystem.”

AI fraud detection is already in place in parts of the financial sector. Resistant.AI, for instance, offers know-your-customer fraud detection that analyses documents, including invoices and proof of identity for KYC, to find alterations that may be missed by a human.

The use of AI is growing across the trade finance sector. Last week Complidata signed a deal with ABN Amro to help staff analyse letters of credit and documentary collections, following closely on from AI compliance checking provider’s partnership with Surecomp.

Others in the industry hope that AI can be used to increase access to credit for SMEs by helping to simplify credit risk analysis.