A Security Architecture for Multi-Agent Matchmaking
Leonard N. Foner · 1996
Using a multi-agent system to match and introduce users who share share interests has important advantages, but handling sensitive data involves a number of design challenges in ensuring user privacy. This paper describes many of them, briefly summarizes some relevant cryptographic technology, and uses this technology to demonstrate how to avoid most of the potential privacy problems without unacceptable performance penalties. INTRODUCTION In general, agents are useful because they understand some aspects of their users' goals and can carry out actions autonomously to fulfill those goals. This may require that any given agent know personal or sensitive information about its user, and that it must be robust against revealing this information to third parties or allowing its actions to be subverted by a malicious interloper into carrying out an undesired action. As the value of the information or the motivation to inflict damage increases, the possibility of inadvertent information dis...