- 1 - ARE ANONYMOUS AGENTS REALISTIC? Frances Brazier 1 , Anja Oskamp 2 , Corien Prins 3 , Maurice Schellekens 3 and Niek Wijngaards 1 1 Intelligent Interactive Distributed Systems, Faculty of Sciences Vrije Universiteit Amsterdam, de Boelelaan 1081a, 1081 HV, Amsterdam, The Netherlands Email: {frances, niek}@cs.vu.nl Phone: +31 - 20 - 444 7737, 7756; Fax: +31 - 20 - 444 7653 2 Computer and Law Institute, Faculty of Law Vrije Universiteit Amsterdam, de Boelelaan 1105, 1081 HV, Amsterdam, The Netherlands Email: a.oskamp@rechten.vu.nl Phone: +31 - 20 - 444 6215; Fax: +31 - 20 - 444 6230 3 Center for Law, Public Administration and Informatization, Faculty of Law Tilburg University , P.O. Box 90153, 5000 LE, Tilburg, The Netherlands Email: {J.E.J.Prins, M.H.M.Schellekens}@uvt.nl Phone: +31 - 13 - 466 - 8044; Fax: +31 - 13 - 466 8149 Abstract. Software agents are involved in Internet applications such as E-commerce and may contain identificatory information about their human user such as credit cards and bank accounts. This paper discusses whether human users and software agents are allowed to be anonymous and whether anonymity is technically realisable from the perspective of Artificial Intelligence, Computer Systems and Law. 1. Introduction In daily life, many situations occur in which humans may be anonymous. For example, a human buyer may wish to be temporarily anonymous during an auction by employing a middleman, as to not influence the price. At some (later) moment in time he may need to reveal his identity to complete the transaction. With the advent of the Internet, it becomes possible for software agents to electronically make transactions, e.g. at electronic auctions. Similar to the previous example, there may be situations in which agents may wish to remain anonymous, for example when searching the Internet for medical information. One way to accomplish this is to use middle agents (Wong and Sycara, 2002), but other options exist. In general , there is not much known about the legal status of agents and their actions (e.g., Brazier, Kubbe, Oskamp and Wijngaards, 2002). There is even less knowledge about anonymous agents specifically: are agents, for example, allowed to be anonymous, and is this feasible? These issues are closely related to legal and technical identities of human counterparts of agents: owners, users, deployers, organisation, designer, producers, etc. In the near future legal and technical decisions with respect to these issues will greatly influence the use of software agent applications and -support. If agents have no legal status they will never be used in practice. In theory, anonymity holds for both agents and humans. Human anonymity can be arranged by an agent keeping identificatory information about humans confidential. In specific situations it may be desirable to have agents be ’anonymous’, i.e. not be (easily) traceable in their actions, nor (easily) relateable to a human owner or user. To this end, an agent may strategically employ a number of identities: pseudonyms. This implies that various levels of anonymity exist, ranging from brief temporal anonymity to absolute