We develop an approach to security of adaptive agents that is based on respecting the local security policies of agents rather than imposing a global security policy on all agents. In short, an agent can be assured, that it will not be able to observe any violation of its own security policy due to the changing presence of other agents in its environment. The development is performed for a version of Dijkstra's Guarded Commands with relocation primitives, channel based communication, and explicit non-determinism. At the technical level a type system enforces local security policies whereas a reference monitor ensures that relocation is permissible with local security of all agents.