AI-Driven Cybersecurity Risk Management for AI-Powered Business Systems

An integral part of current business systems, Artificial Intelligence is opening up new opportunities for automation, decision support, and intelligent service delivery but also offers new cybersecurity concerns. This research aims to explore the nature of cyber threats targeting AI and LLM environments by analyzing a secondary AI Sec Watch Dataset. The research explores threat severity, threat exploitability, sophistication of attack, targeted AI functions, and threat prioritization to determine meaningful cybersecurity threat patterns. A qualitative research design is used to analyze 3020 threat records and the related security attributes, backed up with quantitative exploratory data analysis. Key threat variables are explored in terms of their distribution and relationships by descriptive statistics and visual analysis. This analysis has shown that medium and high-severity threats are the most common recorded events, which shows that there is a high level of exposure in AI-powered environments. The analysis also illustrates that vulnerabilities of the same severity can have different probabilities of being exploited, thus depending on the number of risk factors assessed. AI risks are not evenly distributed among AI components like large language models, agents, APIs or vector databases, as there are differences in sophistication of attack and concentration of threat. Furthermore, there is a strong correlation between threat prioritization patterns and threat severity, exploitability and impact. The study helps to enhance the understanding of AI cybersecurity threats and offers practical implications that may assist in the risk assessment, security planning, and prioritization process for AI-powered business systems.