Cyber security is a great challenge for organizations. Big Data and Machine Learning can be used to improve cyber security protection mechanisms for organizations. This article presents two applications of Latent Dirichlet Allocation (LDA) model for cyber defense. In the first application, we present how to use LDA to discover users' possible hidden intentions from a collection of the user's operations extracted from a very large volume of security monitoring data. This new application can help intrusion detection and malicious activities conducted by insiders. In the second application, we present an architecture used for identifying and redacting sensitive information from outgoing emails in an organization, which may help an enterprise or organization to protect their sensitive information from both intended or unintended leak.