Abstract
The development of IIoT ecosystems and smart city infrastructure leads to the complexity of digital systems and the expansion of cyber capabilities. Traditional pitting techniques are unable to provide the necessary automation and scaling in dynamic, adaptive and hybrid threats. In response to these challenges, the article proposes a hybrid methodology combining human importance with the capabilities of AI-agents, LLM and ML. The proposed methodology provides intelligent information gathering, scanning and modeling of attacks under AI basis, verification of AI-models' resistance to threats and formal risk assessment. The study studied in detail the processes, methods, tools and principles of AIenhanced penetting, as well as the importance of human. Presented new methodology AI-Pentesting and complex taxonomy. The proposed approach could be a further inspiration for research on peneting in the security of IIoT infrastructure and smart cities.