Exercise 1: The Injection Scenario

Think through a concrete scenario in which prompt injection could affect you specifically. Consider the AI tools you use that process content from external sources — emails, documents, web pages, user inputs from others.

For one of those tools, describe what a prompt injection attack might look like. What would the attacker be trying to accomplish? What hidden instruction would they embed in the content your AI processes? What would the consequence be if the injection succeeded?

This exercise is about developing the mental model, not about finding a real vulnerability. The goal is to understand how prompt injection connects to your actual workflow and what it would mean if it happened to you. Write your scenario in enough detail that you can visualize it.

Example structure: "I use [AI tool] to [specific task]. An attacker could craft [specific content type] containing [hidden instruction]. If my AI followed that instruction, it would [consequence]." But go deeper than the example — make it specific to your actual usage.