Communicating security agents

The paper discusses the potential effectiveness of ubiquitous, communicating, dynamically confederating security agents for monitoring and controlling communications among the components of preexisting applications. These agents remember events, communicate with other agents, draw inferences, and plan actions to achieve security goals. Key features of this approach are: (1) linguistic mechanisms for specifying agents, security models, and communications; (2) compilation mechanisms that automatically create and install agents as wrappers around existing application components; (3) algorithmic definitions of how agents communicate to increase the security of systems; and (4) a library of agent code fragments out of which the compilation mechanism builds actual agents. Automating the generation of security agents raises the possibility of the cost effective generation of enough redundant agents to tolerate some erroneous or subverted elements