What security risks should be considered when deploying autonomous agents?