A security framework for agent-based systems

Purpose – This paper aims to address some security issues in open systems such as service-oriented applications and grid computing. It proposes a security framework for these systems taking a trust viewpoint. The objective is to equip the entities in these systems with mechanisms allowing them to decide about trusting or not each other before starting transactions. Design/methodology/approach – In this paper, the entities of open systems (web services, virtual organizations, etc.) are designed as software autonomous agents equipped with advanced communication and reasoning capabilities. Agents interact with one another by communicating using public dialogue game-based protocols and strategies on how to use these protocols. These strategies are private to individual agents, and are defined in terms of dialogue games with conditions. Agents use their reasoning capabilities to evaluate these conditions and deploy their strategies. Agents compute the trust they have in other agents, represented as a subjective quantitative value, using direct and indirect interaction histories with these other agents and the notion of social networks. Findings – The paper finds that trust is subject to many parameters such as the number of interactions between agents, the size of the social network, and the timely relevance of information. Combining these parameters provides a comprehensive trust model. The proposed framework is proved to be computationally efficient and simulations show that it can be used to detect malicious entities. Originality/value – The paper proposes different protocols and strategies for trust computation and different parameters to consider when computing this trust. It proposes an efficient algorithm for this computation and a prototype simulating it.