Route selection in cognitive packet networks (CPNs) occurs continuously for active flows and is driven by the users' choice of a quality of service (QoS) goal. Because routing occurs concurrently to packet forwarding, CPN flows are able to better deal with unexpected variations in network status, while still achieving the desired QoS. Random neural networks (RNNs) play a key role in CPN routing and are responsible to the next-hop decision making of CPN packets. By using reinforcement learning, RNNs' weights are continuously updated based on expected QoS goals and information that is collected by packets as they travel on the network experiencing the current network conditions. CPN's QoS performance had been extensively investigated for a variety of operating conditions. Its dynamic and self-adaptive properties make them suitable for withstanding availability attacks, such as those caused by worm propagation and denial-of-service attacks. However, security weaknesses related to confidentiality and integrity attacks have not been previously examined. Here, we look at related network security threats and propose mechanisms that could enhance the resilience of CPN to confidentiality, integrity and availability attacks. © 2014 Inderscience Enterprises Ltd.