Security and inference in multilevel database and knowledge-base systems

This paper addresses the threat to multilevel security that arises from logical inference and the semantics of the application. Such compromises of security are particularly challenging since they circumvent traditional security mechanisms and rely on a user's knowledge of the application. The problems of inference and security have heretofore been amorphous and difficult to circumscribe. We focus on these problems in the context of a multilevel database system and show their relevance to knowledge-based systems, sometimes referred to as expert systems. Here we establish a framework for studying these inference control problems, describe a representation for relevant semantics of the application, develop criteria for safety and security of a system to prevent these problems, and outline algorithms for enforcing these criteria.