Automated generation of enforcement mechanisms for semantically-rich security policies in Java-based multi-agent systems

Policies are being increasingly used for controlling the behavior of complex systems (including agent systems). The use of policies allows administrators to specify agent permissions and obligations without changing source code or requiring the consent or cooperation of the entities being governed. Past approaches to policy representation have been restrictive in many ways. By way of contrast, semantically-rich policy representations can reduce human error, simplify policy analysis, reduce policy conflicts, and facilitate interoperability. However, semantically-rich policies increase the complexity of fielding policy-governed multi-agent systems. This paper discusses some technical challenges to automatically enforce semantically-rich security policies in Java-based multi-agent systems and presents an engineering approach for addressing some of these challenges. We have developed a first implementation that allows to enforce OWL policies represented using the KAoS policy framework into multi-agent systems built on top of the JDK1.4. The proposed solution allows to control the behavior of agents at a high level of abstraction and exploits the security mechanisms provided by the Java authentication and authorization service (JAAS) to enforce OWL policies.