The robotic process automation (RPA) in enterprise computing refers to software bots that are capable of mimicking most of the human�computer interactions to carry out day-to-day business operations. This RPA is targeted for automating rule-based repetitive and high-volume tasks with higher accuracy which eventually reduces the operation cost and processing time. The cognitive automation which is being developed as part of enterprise computing automation best utilizes Artificial Intelligence (AI) and Machine Learning (ML) techniques to enhance the process model of RPA in terms of improved accuracy, reliability, and consistency in taking intelligent business decisions with no or minimal human interventions. The majority of the communications between computing nodes and bots are being provisioned through service-oriented enterprise computing interfaces, which are built based on Service-Oriented Architecture (SOA). The SOA by itself does not possess any security layer and it defaults to the Open Systems Interconnection (OSI) model for security that is inadequate in this modern era of process automation and interfacing. In this paper, the security concerns of Cognitive RPA for enterprise computing are analyzed, and a novel approach is presented for adding-up the security layer for Cognitive RPA. This is an adaptive approach which works on the predict-prevent-learn pattern for effective proactive security as differed from traditional reactive security, where the Artificial Intelligence (AI) techniques are used for �predict� part for predicting potential security threats, the Artificial Neural Networks (ANN) techniques are applied for �learn� part on unsupervised learning of anticipated security vulnerabilities, and security prevention algorithms are equipped for �prevent� part to defend against the security threats on Cognitive RPA systems. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.