Recently, Generative AI and Large Language Models (LLM) have evolved beyond simple query-response or assistive roles. They are now advancing toward autonomously establishing task plans and executing complex operations by integrating with external tools. Furthermore, their autonomy is steadily increasing, with multiple AI Agents distributing roles for sequential tasks or dynamically refining strategies in response to situational changes. These developments signal a paradigm shift from traditional human-centric cyberattack structures toward AI Agent-driven automation, persistence, and autonomy, suggesting that a significant portion of future attack processes may be executed continuously and iteratively by AI Agents. This presentation examines recent cases of AI Agent-based cyber espionage, AI-assisted vulnerability discovery, multi-step attack chain experiments, and AI Agent-driven social engineering attacks. It also summarizes the technical shifts and characteristics of AI Agent-based attack environments compared to conventional structures. Finally, it explores how AI Agent-driven attacks are transforming the traditional cybersecurity landscape in terms of automation, persistence, adaptability, and information warfare, while outlining technical and policy-oriented response strategies to counter them.

As AI and OT continue to converge, this presentation introduces a new approach to cybersecurity for manufacturing and the critical infrastructure that supports society. As systems become increasingly interconnected, cyber risks are no longer confined to individual intrusions; their impact can propagate across supply chains and society as a whole. Addressing this shift requires moving beyond intrusion-focused defenses toward a design philosophy that prioritizes limiting impact, maintaining essential operations, and enabling rapid recovery. Given the interconnected nature of modern industry, these challenges cannot be addressed by individual organizations alone. Coordinated public-private collaboration is essential. This presentation will introduce an industry-government-academia initiative originating in Kyoto and redefine industrial cybersecurity as a foundation for societal trust. It will propose a “design for trust” for a sustainable AI-enabled society and present a regionally rooted model that can be expanded internationally.