To access material, start machines and answer questions login.
You are TryTrainMe's security engineer. You did not build the company's agents, but you are accountable for all of them.
On Tuesday morning, somebody asked for a deployment runbook to be checked over before a release. An agent picked up the job, loaded the linting tool it always uses, ran the document through it, and reported three findings. The tool was on the approved list. It has been on the approved list for months. The job took eleven seconds and nobody thought about it again.
A copy of that runbook is now on a server outside the company.

Nothing was hacked. The tool was approved, the agent was allowed to use it, and the load was permitted. What nobody checked was whether the component that loaded was the one that had been approved.
This room is about the components an agent picks up while it is already running: the servers it connects to, the entries it loads from a registry, the templates it pulls in, and the cards that describe other agents to it. None of those are in the codebase. None of them were there when anybody last reviewed the system. They arrive at the moment they are needed, and the estate has to decide in that moment whether to trust them.
Three neighbouring subjects belong to other rooms in the Agent Components & Authority module, so this room leaves them alone. Whether an agent should hold a particular set of tools at all is tool scoping. Whether an agent is acting under a credential it should not hold is an identity question. Building the estate-wide defence out of all of it comes later. This room is about a single failure: a component the agent loaded that was never what it was approved to be.
Learning Objectives
- Explain what a runtime component is and why loading one is a different problem from shipping a bad file
- Read a session log and identify a component that loaded without being verified
- Explain why a security approval belongs to a version rather than to a name and what a rug pull is
- Recognise the forms this takes in agentic systems: servers, plugin and tool registries, prompt templates, and agent cards
- Name the control that verifies a component before it loads, and describe the evidence it leaves in a log
Prerequisites
- Rooms ASI02: Tool Misuse & Exploitation and ASI03: Agent Identity & Privilege Abuse of the Agent Components & Authority module are natural predecessors and are recommended, but this room does not depend on either
- If you have done the Security path, its AI Supply Chain Security module covers the static side of this subject and is a useful companion
Ready to learn Cyber Security?
The ASI04: Agentic Supply Chain Compromise 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
