Inference of sensitive information in data integration systems

September 28, 2015

Mehdi Haddad (LACL)

The inference problem, in an access control context, refers to the ability of a malicious user to synthesize sensitive information from a combination of non sensitive information. This problem is highlighted in data integration systems where a mediator providing a unique entry point to several heterogeneous sources is defined.

In this talk we describe an incremental methodology able to tackle the inference problem in a data integration context. This methodology has three phases. The first phase, the propagation phase, allows combining source policies and therefore generating a preliminary policy at the mediator level. The second phase, the detection phase, characterizes the role of semantic constraints in inducing inference about sensitive information. We also introduce in this phase a graph-based approach able to enumerate all indirect access that could induce accessing sensitive information. In order to deal with previously detected indirect access, we introduce the reconfiguration phase which provides two solutions. The first solution could be implemented at design time. The second solution could be implemented at runtime.