Short Abstract for 'Dynamic Security in Communication Systems' (ISBN 3-8265-6809-5) As of now, the provision of security (confidentiality, integrity, and authenticity) is an all-or-nothing issue. There are no mechanisms to fine-tune the strength of offered security and have applications use just the right amount to deter attackers. In this thesis, dynamic aspects of security in communication systems are explored. This covers Quality of Service (QoS) models and requirements for security, secure multimedia protocols, and peer and component authentication. Each of these issues is examined, and their dynamic properties are shown. The experimental platforms Da CaPo and WaveVideo are used to prototype some of the results. In essence, it is shown that it is possible to provide fine-grained, scalable security to applications and allow them to select at runtime the required amount of processing overhead necessary to achieve sufficient security. One important contribution is the provision of dedicated security QoS parameters. To this end, the new parameters are quantified by establishing a relationship to the monetary value of data. The special case of zero-cost attacks is also presented. Confidentiality and authenticity are modelled from the user's perspective, and translated down to the communication infrastructure. There, QoS parameters are used to select protocol and algorithm properties. Functionality and performance of a prototype have been evaluated, showing that fine-grained control over security is feasible, and the resulting performance is sufficient for real-world applications. Another important contribution is the extension of multi-media protocols to provide dynamically configurable security functionality. In contrast to solutions that operate on a multimedia data stream as a whole, dedicated security mechanisms that are embedded in the rest of the multimedia protocol allow more efficient processing. They receive QoS parameters to control their operations, and then provide adequate confidentiality and authenticity for the transferred data. By varying the employed algorithms and the actual data coverage of the algorithm, computing power consumption can be reduced. This property is critical when real-time behaviour is needed in software-only solutions. It is studied in the context of general purpose, audio, and video data. Results derived from a prototype are presented.