In response to contemporary security challenges for Internet of Things systems, this paper introduces an architectural framework for data driven security monitoring and automation. The architecture supports advanced data analytics for detecting anomalies at all layers of an IoT system, based on a powerful mechanism of reusable security templates. Also, the paper provides a concrete example of data-driven IoT security for smart objects, based on the use of deep learning algorithms and their implementation over the introduced architecture framework. The algorithms are successfully deployed and used for effective and predictive detection of anomalies and abnormalities at the network and application layers of the respective IoT systems. They manifest how deep learning and AI techniques can be used for efficient security in conjunction with the introduced framework.