Financial Data Exchange with Statistical Confidentiality: A Reasoning-based Approach

Abstract

Confidentiality is a crucial requirement in financial data exchange processes. On the one hand, rich microdata is needed for most AI applications, including banking supervision, anti-money laundering, etc. On the other hand, organizations may not be legally authorized to see particular data, e.g., personal data. Striking the right balance provides a number of challenges. Motivated by our experience with the Central Bank of Italy, in this work we present Vada-SA, a reasoning-based framework for financial data exchange with statistical confidentiality. We present a production-ready and fully engineered framework, adopting a reasoning approach. The framework includes explicit consideration of the reasoning process, the business context and declarative transparency that puts the user in control. We show and discuss a number of risk measures and anonymization criteria, implemented and operated in practice