Abstract
Agents often try to convince others to hold certain beliefs. In fact, many network security attacks can actually be framed in terms of a dishonest that is trying to get an honest agent to believe some particular, untrue claims. While the study of belief change is an established area of research in Artificial Intelligence, there has been comparatively little exploration of the way one agent can explicitly manipulate the beliefs of another. In this paper, we introduce a precise, formal notion of a belief manipulation problem. We also illustrate that the meaning of a message can be parsed into different communicative acts, as defined in discourse analysis theory. Specifically, we suggest that each message can be understood in terms of what it says about the world, what it says about the message history, and what it says about future actions. We demonstrate that this kind of dissection can actually be used to discover the goals of an intruder in a communication session, which is important when determining how an adversary is trying to manipulate the beliefs of an honest agent. This information will then help prevent future attacks. We frame the discussion of belief manipulation primarily in the context of cryptographic protocol analysis.,Peer reviewed,Published. Received: 17 January 2014; Accepted: 29 September 2014; Published: 10 October 2014.