Especially in security-critical industries, companies want to ensure that company- internal data remain hidden from unauthorized eyes, e.g., if external workers like cleaning staff are entering critical areas. If an increased security standard is mandatory at the office, it must be ensured that non-authorized persons have no possibility to get access to confidential data, which is displayed on the computer screen. This thesis describes an approach for preserving screen privacy by implementing a context-aware access control framework called Confidential Desktop, which continuously checks whether the user is alone at the office or not. This is done by detecting predefined intruder's activities by the use of different features like detecting faces, motion, sounds and bluetooth devices. The feasibility is shown in simulated scenarios that even with existing standard office hardware and state-of-the-art methods, intruder's activities can be detected by the framework in real-time. Thus the proposed solution can control the access to displayed data and preserve the confidentiality of sensitive information from prying eyes.