Ich versichere von Eides statt, dass ich diese Dissertation selbständig verfasst und nur die angegebenen Quellen und Hilfsmittel verwendet habe. Datum i/ix Abstract Adversaries can deploy rootkit techniques on the target platform to persistently attack computer systems in a stealthy manner. Industrial and political espionage, surveillance of users as well as conducting cybercrime require stealthy attacks on computer systems. Utilizing a rootkit technique means, that a part of the implemented attack code is responsible for concealing the attack. Attack code that is loaded into peripherals such as the network interface card or special micro-controllers currently are the peak of the evolution of rootkits. This work examines such stealthy peripheral-based attacks on the host computer. Peripherals have a dedicated processor and dedicated runtime memory to handle their tasks. This means that these peripherals are essentially a separate system. Attackers benefit from this kind of isolation. Peripherals generally communicate with the host via the host main memory. Attackers exploit this fact. All host runtime data is present in the main memory. This includes cryptographic keys, passwords, opened files, and other sensitive data. The attacker only needs to locate such data. Subsequently, attackers can read and modify the data unbeknownst by utilizing the direct memory access mechanism of the peripheral. This allows for circumventing security software such as state-of-the-art anti-virus software and modern hardened operating system kernels. Detecting such attacks is the goal of this work. Stealthy malicious software (malware) that is based on an isolated micro-controller is implemented to conduct an attack analysis. The malware proof of concept is called DAGGER, which is derived from Direct memory Access based keystroke code loGGER. The development and analysis of this malware reveals important properties of peripheral-based mal-ware. The results of the analysis are the basis for the development of a novel runtime detector. The detector is called BARM-Bus Agent Runtime Monitor. This detector reveals stealthy peripheral-based attacks on the host main memory by exploiting certain hardware properties. A permanent and resource-efficient measurement strategy ensures that the detector is also capable of detecting transient attacks. Such transient attacks are possible when the applied measurement strategy only measures at certain points in time. The attacker exploits this measurement strategy by attacking the system in between two measurements and by destroying all attack traces before the system is measured. The detector represents an alternative solution for previously proposed preventive protection approaches, i. e., input/output memory management units. Previously proposed approaches are not necessarily effective due to practical issues. This fact as well as the threat posed by peripheral-based mal-ware demand the alternative detector solution that is presented in this work. The detector does not only reveal an attack, but al...