This paper proposes a multi-agent and multi-level architecture to secure highly distributed and heterogenous systems, based on an original security meta-policy approach. The resulting system guarantees global security properties with strong fault tolerance abilities. The architecture is divided into several levels, each one contributing to the automation of the security management. This multi-agent based system allows the distributed management of access control, intrusion detection systems, security testing and verification of security policies. Using cooperative agents makes it possible to locally configure the system differently that on other nodes. The meta-policy paradigm guarantees local evoltuion of each local policies in conformance with the meta security properties. The verification components guarantees that global evolution of the meta-policy maintains it consistent. Test components provides interesting abilities such as working metrics auto evalution. It can also evaluate the existing IDS our systems uses or our meta-policy based IDS. The test components also provides internal security metrics in testing the corruption or the well working of our agents.