A Sanctuary for Mobile Agents

The Sanctuary project at UCSD is building a secure infrastructure for mobile agents, and examining the fundamental security limits of such an infrastructure. First, what do we mean by \secure"? An obvious issue is the privacy of computation. With standard approaches for agent-based systems, a malicious server has access to the complete internal state of an agent: software agents have no hopes of keeping cryptographic keys secret. The privacy of computation is only one aspect of the security picture: the integrity of computation is perhaps more critical. In agent-based computing, most researchers have been concentrating on one side of the security issue: protecting the server from potentially malicious agents. Related work in downloadable executable content (Java [7], Software Fault Isolation [19], Proof-Carrying Code [16, 17], OS extension mechanisms such as packet lters [13], type safe languages [4, 9], etc) all focus on this problem. The converse side of the agent security problem, however, is largely neglected and needs to be addressed: how do we protect agents from potentially malicious servers? Why should we believe that the result returned by our software agents are actually correct and have not been tampered with?