On the challenges of formal privacy models for open banking

In recent years, financial data sharing systems have been rolled out in multiple countries under the umbrella of “Open Banking” or “Open Finance”, to facilitate the sharing of financial information. It is usually the case that financial information is shared in the form of microdata, which is known to be vulnerable to attacks such as re-identification and attribute inference. Previous work shows that this is the case even if data is de-identified, and even if no personal information is available, and it gets substantially worse the more data is shared by customers. It does not, however, evaluate the effectiveness of privacy mechanisms applied to the transaction data. Although this type of analysis has been formalised before, previous approaches targeted only syntactic (deterministic) privacy mechanisms, and are not generalisable to other complex scenarios. In this work, we describe current challenges in formally modelling open banking scenarios. In particular, we: (i) propose a novel formal model of linkage attacks against privacy-preserving data-release systems; (ii) describe the connection between privacy analyses for data collection and dynamic leakage models, which currently has no accepted definition that produces sound results; and (iii) present our recent advances on a novel formalisation of dynamic leakage which satisfies sound information-theoretic properties.