Identifying private information and preventing privacy violations is provided by a process that evaluates digital information obtained by an organization as part of a digital information stream from a user. The evaluating identifies a user utterance, including keyword(s), entity/ies, and intent(s), and applies natural language understanding to the digital information to ascertain a contextual understanding for the user utterance. The process selects training set(s) of historical information from available training sets that includes vocabulary used in varying contexts. The process compares the identified user utterance to an ontology based on the selected training set(s), and determines a confidence level that the digital information includes digital private information. The process also flags for the organization an action to take with respect to handling of the digital information. The flagging is based on the determined confidence level that the digital information includes digital private information.