The conceptual architecture of the access control system described here is based on automatic distributed acquisition and processing of knowledge about users and devices in computer networks. It uses autonomous agents for distributed knowledge management. Agents grouped into distributed communities act as mediators between users/devices and network resources. Communicating with each other, they make decisions about whether a certain user or device can be given access to a requested resource. In other words, agents in our system perform user/device authentication, authorisation, and maintenance of user credentials.