A Solution for Securing the Information Environment Inspired by Living Organisms and Biology

Abstract

Cyber-attacks have become increasingly frequent and severe in recent years, posing a significant threat to both economic and safety domains. In order to mitigate these threats, it is crucial to have a thorough understanding of the various types of attacks, including their behaviour and properties. Bio-cyber operation is a new field of research that draws inspiration from the human immune system, which has found solutions to problems that cybersecurity professionals have been struggling with for decades. By studying the human immune system, cybersecurity can teach valuable lessons on how to detect and deter attacks. To address the issue of sensitive information or data leakage, a "cyber immune" technology can be employed to detect unknown cyber-attacks and provide a powerful defence mechanism. This thesis proposes a methodology for predicting potential vulnerabilities by utilizing natural language processing (NLP) and analysing "Common Vulnerabilities and Exposure" (CVE) reports. Additionally, the implementation of Gradient Sensitivity (GS) and Gradient Sensitivity Input (GI) techniques into our framework represents a significant leap forward in enhancing the model's explainability. The incorporation of GS and GI techniques into our framework means that our models meet the critical needs of cybersecurity applications today. They strike a balance between advanced predictive capabilities and the equally important need for transparency and interpretability. This balance is what will keep driving cybersecurity efforts forward, ensuring that we're not just creating models that can predict threats but also fostering a cybersecurity approach that can effectively act on these predictions. Finaly, by examining the field of human biology, we can gain significant insight into the bio-cybersecurity domain, which can be applied to developing more effective cybersecurity measures. Our results indicate that by using NLP we can achieve high percentage of accuracy to predict the possible vulnerability (more than 94%) as well as find the interrelationship between different vulnerabilities (more than 53% similarity).