Towards Using Possibilistic Information Flow Control to Design Secure Multiagent Systems

We show how security requirements, in particular confidentiality requirements, for a whole multiagent system can formally be decomposed into confidentiality requirements for the agents. The decomposition assumes that there is some control over, or trust in, a subset of the agents and that the platform is trusted to satisfy certain reasonable assumptions. It is generic over the internal execution model of the agents. The decomposition is carried out in full detail for one specific class of confidentiality requirements, yielding a theorem that can be directly applied to derive confidentiality requirements for single agents from the overall requirement. Similar decompositions for other global requirements or under slightly different assumptions about the platform can be carried out along the same lines.For expressing security requirements we use an existing framework for possibilistic information flow control, profitting from, e.g., the framework’s available composition results. The decomposition, because it is carried out formally and rests on a well-studied framework, is fully rigorous and the resulting property of the overall system is well-understood.