Abstract
Within agent systems, two entities, namely hosts and agents, subsist and require protection against potential malicious acts. The use of such agent systems to support the development of practical applications is limited primarily by the risks to which hosts in the system are subject to. This article introduces a distributed and adaptive security-monitoring framework to decrease such potential threats. The proposed framework is based on a modified version of the popular Boosting algorithm to classify malicious agents based on their execution patterns on current and prior hosts. Having implemented the framework for the Aglet platform, we herein present the results of our experiments showcasing the detection of agent entities in the system with intention deviating from that of their well-behaved counterparts.
