What you'll learn

Autonomous AI agents take actions; they call tools, read external content, and make decisions. That creates an attack surface far larger than a simple conversational chatbot has. Prompt injection allows an attacker to hide instructions inside content an agent processes, and goal hijacking, gets the agent to abandon its intended task to follow them instead. In this course, Agentic AI Security: Prompt Injection and Goal Hijacking, you'll gain the ability to identify these attacks as they happen and design agents that limit the damage when they do. First, you'll explore why Agentic AI carries a fundamentally different risk profile than conversational LLMs, and how direct and indirect injection can start to control what an agent does. Next, you'll discover how to detect deviation in workflows, gathering enough information to assess what went wrong, where it went wrong, and use that information to bolster your defenses. Finally, you'll learn how to design attack-resistant agents using a variety of logical and fuzzy controls, layered together so that no single point of failure can compromise the whole system. When you're finished with this course, you'll have the skills and knowledge to protect your workflows from prompt injection and goal hijacking attempts.