METHOD FOR ASSESSING CYBER INCIDENTS WITH ARTIFICIAL INTELLIGENCE

Abstract

The expanding use of artificial intelligence in information systems, cloud platforms, automated processes, and autonomous agents changes the characteristics of contemporary cyber incidents. Artificial intelligence may simultaneously become a target of attack, a tool supporting malicious activity, an autonomous participant in a sequence of actions, or a factor accelerating and propagating an incident. This limits the applicability of traditional assessment approaches that do not explicitly account for autonomy, access to external tools, loss of human oversight, and propagation between interconnected AI components.
This paper proposes a method for classification and quantitative assessment of cyber incidents related to artificial intelligence. The method first determines the role of AI in the incident and then evaluates its technical severity through normalized indicators. Particular attention is given to the degree of autonomy, propagation potential, impact on data and information resources, privilege abuse, and the level of retained human oversight. An integrated score is introduced by combining the baseline incident severity with corrections reflecting the joint influence of autonomy and propagation and the loss of human control. The method is intended to be examined through scenarios involving AI-assisted phishing, prompt injection, compromise of an autonomous AI agent, and propagation in a multi-agent environment.