A system prevents divulgation of sensitive data in two snapshots, taken at different times, of one or more systems. The system identifies a set of files from among file pairs. Each file pair is formed from a respective file that includes a difference with respect to each of the two snapshots. The system performs a pattern reducing process that removes, from the set, any file having, as the at least one difference, a predetermined non-sensitive difference between respective executions of a pre-determined system operation. The system performs a commonality reducing process that removes, from the set, any file having, as the at least one difference, a common difference between different system users. The system annotates data in remaining files in the set as potentially being sensitive data. The predetermined non-sensitive difference is determined using a Sandbox host. The common difference is determined using an actual one of the systems.