Ph.D. candidate in Ingegneria Informatica E Dei Sistemi , 41st cycle (2025-2028)
Department of Control and Computer Engineering (DAUIN)
Profile
PhD
Research topic
Agentic AI for Cybersecurity: Autonomous RedBlue Agents in Interactive Cyber Environments
Tutors
Keywords
Biography
My work focuses on the design and evaluation of autonomous red/blue/purple teaming systems multi-agent frameworks in which a red team agent continuously attempts to compromise a target environment, while a blue team agent simultaneously works to detect, contain, and respond to those attacks. The two agents are coupled in a contrastive learning loop: every successful attack teaches the defender a new weakness to patch, and every successful defense forces the attacker to discover a more sophisticated strategy. Over time, this adversarial dynamic drives both agents toward increasingly expert behavior without requiring human intervention at each step.
My research contributes to several open problems at the boundary of AI and security: how to design reward functions that capture meaningful security outcomes, how to build environments that faithfully simulate realistic attack surfaces, and how to ensure that agents trained in simulation transfer usefully to real-world settings. I am particularly interested in the role of large language models as reasoning engines within these agentic pipelines building on my earlier work in LLM security to understand both their potential and their failure modes in adversarial contexts.
Ultimately, my goal is to lay the groundwork for a new generation of AI-driven security tools: systems that do not merely automate known playbooks, but continuously discover, adapt, and evolve.