Back to all walkthroughs
ASI01: Agent Goal Hijack
See how one poisoned email let EchoLeak steer Copilot, and why an agent's goal is so easy to steal.
medium
To access material, start machines and answer questions login.
The Agentic Fundamentals module has taken you through everything you need to know to reframe your mindset to secure agentic technology: the fundamentals of how the technology works, how we go about threat modelling such an advanced and unpredictable technology, and how we monitor it. From this module, Agent Hijacking, onwards, we will take a more in-depth look at each of the Top 10 vulnerabilities for Agentic Applications. Starting, naturally, with ASI01: Agent Goal Hijack. Let's go in depth and get hands-on!
Prerequisites
- It is recommended that you have completed the Agentic AI Fundamentals module of the Agentic AI Security path.
- A good grasp of pre-existing attacks, such as those covered in the Prompt Injection and Jailbreaking rooms.
Learning Objectives
- Define ASI01 (Agent Goal Hijack)
- Explain how ASI01 differs from LLM01 (Prompt Injection)
- Explain why a model can't reliably separate instructions from content
- Identify what makes an attack zero-click, and recognise ASI01 in the EchoLeak attack
- Understand ASI01 mitigations
Answer the questions below
I'm ready to go!
Ready to learn Cyber Security?
The ASI01: Agent Goal Hijack room is only available for Premium or Max subscribers. Signup now to access more than 500 free rooms and learn cyber security through a fun, interactive learning environment.
Already have an account? Log in