Beyond Models: Building Trusted AI in the Age of Autonomous Agents

Abstract

This presentation introduces the emerging paradigm of multi-agentic AI in cybersecurity, where systems evolve from passive tools to autonomous decision-making entities. This explores the rise of agentic AI and its transformative impact on modern cybersecurity. As AI systems evolve from assistive tools into autonomous agents capable of independent decision-making and action, they introduce a new class of dynamic, fast-moving, and highly adaptive threats. The session examines the architecture and behavior of these agents, highlighting critical vulnerabilities that emerge from autonomy, orchestration, and trust dependencies. It presents the concept of a "lethal trifecta" of risks that redefine traditional security boundaries. Through practical scenarios, the workshop illustrates how adversarial agents exploit interconnected systems and data flows at scale. It then outlines a defense-first approach grounded in observability, governance, and secure infrastructure design, emphasizing a shift toward "secure-by-design" principles. The discussion also addresses the evolving balance between human oversight and machine autonomy. Participants will gain actionable frameworks to design resilient, trustworthy agentic systems and transition toward proactive, self-healing cybersecurity models.