Data Leakage Detection

A data distributor provides confidential data to a team of reliable agents (third parties). Even so, some of this data ends up in the wrong hands, for example listed on websites or found on people’s computers. It is the distributor’s task to decide if the data likely came from agents, rather than being found by more general means. This work introduces data allocation methods across agents to raise the likelihood of early data leak detection. No changes such as adding watermarks, are necessary to these strategies for released data. We also look at how synthetic but genuine data can improve our chances of tracing the leak. The aim is to find out when private data is leaked by an agent and to spot this agent if identification is feasible.