A Data Loss Prevention (DLP) system includes an automated method for tracking changes to a security classification (e.g., content category) associated with an artifact to determine whether an attempt is being made to subvert a DLP policy. The method exploits the basic principle that, depending on context, the classification of a particular artifact, or a change to an existing classification, may indicate an attempt to subvert the policy. According to the method, an artifact classification state machine is implemented within a DLP system. For each policy-defined content category on each artifact, the machine identifies a content category change that may be of interest, as defined by policy. When a change in a classification has occurred, an artifact notification event (or, more generally, a notification of the change in classification) is issued.