Preserving confidentiality while reacting on iterated queries and belief revisions

Abstract

In multiagent systems, agents interact and in particular exchange information to achieve a joint goal, e.g., arrange a meeting, negotiate a sales contract etc. An agent, as a rational reasoner, is able to incorporate new information into her belief about her environment (belief revision) or to share her belief with other agents (query answering). Our agent model is based on a common line of research where belief revision is seen as the process of nonmonotonic reasoning from the available information. Yet, such an agent might be interested to hide confidential parts of her belief from another requesting agent and, thus, must control the respective reaction to a revision or query request. As our first contribution, we define the confidentiality aims of the reacting agent and postulate the requesting agent's capabilities in attacking these interests. In particular, we study an operator by means of which the requesting agent attempts to skeptically entail confidential beliefs of the reacting agent from observed reactions. This skeptical entailment operator is based on a class of nonmonotonic consequence relations such that the reacting agent's reasoning is implemented as an instance of this class. As our second contribution, we give an algorithmic solution for the reacting agent to enforce her confidentiality aims. To this end, we show how skeptical entailment could be computed via deduction with respect to an appropriate axiomatization of the class of consequence relations on which skeptical entailment is based. In particular, we present control procedures using the skeptical entailment operator and prove that these procedures effectively enforce confidentiality by means of refusal even if the requesting agent also takes their execution into consideration (meta-inference).