Robust AI Security and Alignment: A Sisyphean Endeavor?

About this session

The industry is rapidly transitioning from static AI chatbots to Agentic AI. Unlike their predecessors, these agents do not merely offer advice; they execute actions via APIs across your core IT infrastructure to automate complex workflows. This shift fundamentally expands the attack surface and renders the traditional "cybersecurity perimeter" obsolete.At the heart of these agents lies the Large Language Model (LLM)—a component that remains stubbornly vulnerable. It is now a predictable cycle: a frontier model is released, hardened, and subsequently "jailbroken" within days. This is not a failure of engineering, but a reality of mathematics. Rigorous theory suggests that no finite set of guardrails is universally robust; there will always exist a prompt capable of evading a model’s defenses.If the core is inherently fragile, how do we secure the autonomous future?In this session, we move beyond the "cat-and-mouse" game of prompt injection to explore the Agentic AI Risk space. We will provide a concrete blueprint for:i) Identifying vulnerabilities unique to agentic tool-use and multi-step execution;ii) Measuring risk within autonomous feedback loops;iii) Architecting resilience for a threat landscape where traditional guardrails are no longer a single point of failure.Don’t manage risk in retrospect. Join us to build a security strategy that thrives in the 2026-and-beyond landscape.