A method, system and computer program product for preventing sensitive pattern disclosures from Linked Data graphs. The proposed method (i) receives as input a Linked Data graph and a set of query patterns that correspond to sensitive knowledge that needs to be concealed, and (b) minimally distorts the Linked Data graph to generate a sanitized counterpart (graph) in which only the non-sensitive patterns can be discovered. The method supports a variety of utility functions, which it optimizes during the graph sanitization process. The resulting, sanitized graph can be subsequently used for publishing and/or querying purposes.