Discover how Microsoft AI secures transactions in banking – A Swift success story
Banking relies on the security and reliability of their institutions and operations. Guaranteeing these principles is then a major priority for all banks. Let’s explore a specific initiative to improve security in banking transactions.
The organization
Swift (Society for Worldwide Interbank Financial Telecommunication) enabled communications between banks and financial institutions since it was founded in Belgium in 1973. The organization sets a standard used all over the world. This way, Swift makes it possible to transfer, receive, and process money and security yearly more than nine billion financial messages.
Swift infrastructure connects over 11,500 financial institutions and delivers services and products in more than 200 countries or regions. Collaboration and synergies are key for Swift culture and success. They’re used to using their vast network of banks to find global solutions to shared problems.
The challenge
Trust and security are the basis of Swift’s business. However, the industry is facing a rise in financial crime, due to the increase of cross-border transactions and instant payment networks. This issue is already costing hundreds of billions annually, including fraud remediation and fund recovery.
The sector demands a solution to fight financial crime effectively. Only a network as large as Swift’s can carry out such a demanding project. Microsoft is also collaborating to power this solution with their platform and AI models.
The solution
Swift decided to build a highly accurate model for anomaly detection to stop fraud. The solution is built in Azure Machine Learning, the Microsoft platform for managing AI systems, and uses Azure Confidential Computing and Microsoft Purview to ensure data privacy.
Swift and Microsoft chose the federated learning technique to build this AI. This approach consists in training the model in independent, decentralized sessions. The advantage of federated learning is that banks participating in the project aren’t required to share training data, because each of them trains the model with their own dataset.
Following this philosophy, Swift developed a first anomaly detection model and shared it with their member banks. Each bank is enriching the model with their own datasets, which increase the accuracy of the resulting models. This workflow is possible because Azure Machine Learning enables you to train a model based on distributed datasets.
The key to this distributed architecture is ensuring data confidentiality. The solution uses Azure Confidential Computing, Microsoft Purview, and a zero-trust-based policy framework that ensures Azure Machine Learning can ingest the distributed datasets without copying or moving data from their secure locations.
The results
Swift is succeeding in building the most accurate anomaly detection model for FSI ever created. This AI will help protect payments all around the world. The solution is already reducing costs in fraud remediation and fund recovery.
To learn more, read Swift innovates with Azure confidential computing to help secure global financial transactions.
Next, let’s discuss a customer story in insurance.