We present an organisation infrastructure for open MASs (Multi-Agent Systems) built upon a role-based access control model (RBAC) allowing agent actions toward the other entities in the MAS (agents and resources) to be enabled and controlled.

To this end, an Agent Coordination Context (ACC) is created by the infrastructure and assigned to an agent as it enters the MAS. An ACC acts as a sort of private interface for the agent toward the environment: any agent action is thereafter enabled and controlled by its own ACC, which prevents those behaviours that are incorrect with respect to the role(s) played by the agent. For this purpose, each role is assigned a policy that flexibly specifies admissible actions and perceptions over time.

Details of this infrastructure are developed formally, through a process-algebraic description of the main infrastructure entities: this is meant to serve as a formal specification of the infrastructure, as well as of the language for expressing policies.