MediumAI SecurityTheory

Agent Security — Exfiltration Attacks on a Coding Assistant

You are the AI security engineer for a company deploying an AI coding assistant that can read and write files in the users project directory, execute shell commands in a container, search internal documentation, and post to a team Slack channel. (1) Describe three specific attack scenarios where indirect prompt injection could cause the agent to exfiltrate sensitive data — for each where is the injection point, what does the injected instruction say, and what tool is used for exfiltration? (2) For each attack describe the capability restriction or monitoring control that would prevent or detect it. (3) The product team wants to add HTTP requests to any URL. What security controls would you require before approving this?

Sign in to attempt this problem

Free account gives you full access to community problems with the complete solution reveal — golden answer, senior walkthrough, and score breakdown — after submission.

Start free →Already have an account? Sign in