As AI systems move from assistance to autonomous action, they introduce a new class of security risk that traditional controls were never designed to handle. When AI acts at runtime - calling APIs, moving data, making decisions - it creates fresh attack paths, new opportunities for data leakage, and failure modes that don’t show up in design-time reviews or policy documents.

This workshop focuses on what actually breaks when AI runs in production. Drawing on real-world scenarios, it examines how AI behaviour can be abused what to watch for at runtime and which technical controls can be applied to detect, contain, and mitigate AI-driven threats before they become incidents.