Agent Trustworthiness

Agent-based technology could revolutionize the manner by which distributed computation is performed. The fact that the information returned by an agent to the agent owner cannot be validated by the owner is impeding the widespread adoption of agent-based computing. Our paper addresses this concern by proposing a new type of software assertion to increase observability by providing agent owner's with agent state \snapshots." These snap-shots provide agent owners with: (1) a means to determine whether its agent's results are trustworthy, (2) information to debug a roving agent, (3) a greater ability to meet real-time constraints, and (4) a means to identify hosts systems that are resource-de cient, grant insu cient access rights, or tamper with agents. We present a methodology and tool for selecting and embedding protective assertions into agent code. We also discuss how the information from the assertions is automatically analyzed. Although our proposed assertions are not foolproof, they make it much harder for an agent to be tampered with in ways that are not detectable by the agent's owner. This knowledge is paramount for the utility of an agent-based system.