Actually, there is growing consensus that for many system applications, safety as well as security demands have to be observed in a coherent manner. In this paper we describe an integrated approach to protect the nodes of distributed co-operative systems against malicious attacks and unplanned system failures. The basic strategy is the use of special diagnostic agents for that purpose. This agent concept is supported by means of additional diagnostic units modularly added to the processor/memory interface of each node of the system. These units have their own autonomous control which cannot be altered by their corresponding processor. Each instruction transferred to the processor, and each data word transferred to/from the processor, in a side step can be scanned by the diagnosis unit. In case of a suspicion for malicious or non-malicious faults, the diagnosis unit can take over control of the corresponding processor to run diagnostic routines, and can trigger recovery procedures to restore a proper state of the processor of the node.

The resulting impact on system reliability is derived; here a reliability modelling approach is discussed to describe especially malicious intrusion faults in a more refined way, by distinguishing different classes of attackers. The resulting reliability of the presented architecture is sketched.