The autonomic paradigm has been introduced in order to cope with the growing complexity of management. In that context, autonomic networks and systems are in charge of their own configuration. However, the changes that are operated by these environments may generate vulnerable configurations. In the meantime, a strong standardization effort has been done for specifying the description of configuration vulnerabilities. We propose in this paper an approach for integrating these descriptions into the management plane of autonomic systems in order to ensure safe configurations. We describe the underlying architecture and a set of preliminary results based on the Cfengine configuration tool.