Agentic AI systems don’t just generate responses — they take actions, call tools, move data, and influence other agents. With this new autonomy comes a new attack surface. OWASP’s 2026 Top 10 for Agentic Applications is the first global framework for securing autonomous AI agents. In this session, we’ll focus on the Critical Four — the four risks that account for most real‑world incidents today: Agent Goal Hijack, Tool Misuse, Identity & Privilege Abuse, and Memory Poisoning. You’ll learn how attackers exploit them, why traditional guardrails fail, and practical mitigations you can apply immediately. If you’re building, deploying, or securing AI agents, this session will show you where the real dangers are — and how to defend against them.”