"Such widespread concerns can only be addressed using micro data and access to this often involves lengthy and cumbersome procedures through review boards and committees for approval, and sometimes it is just not possible."

Dr Boulos, together with colleagues from the Department of Computer Science at the University of Bath and the University of Iowa, suggests that new agent software may be able to overcome some of these problems.

Software agents are programmes that can respond to changes in their environment, generate and attempt to achieve goals, and have the capacity to interact with other agents and even co-operate.

This means that agents could be sent to the original data repository in order to carry out the analysis there, and then send back an aggregate report that does not reveal individual identities. "Software agents could provide flexible but controlled access to unmodified confidential data, and return only results that do not expose any person-identifiable details," said Dr Boulos.

"The use of software agents is not a simple as it sounds, and also carries with it its own security risks, which must be properly addressed.

"Mechanisms need to be introduced that, for example, digitally sign and authenticate genuine agents and their transactions, and prevent 'Trojan horse'-like attacks by fake or rogue agents.

"These mechanisms could include the creation of virtual institutions to insulate host organisations from agents and minimise leakage by limiting access to only the necessary data."

'"/>

Contact: Andrew McLaughlin
a.mclaughlin@bath.ac.uk
44-122-538-6883
University of Bath
7-Sep-2005