Module 09 Knowledge Check
5 questions. Pass mark 4 out of 5. Answer every question before checking the answer key below, then retry after reading the feedback.
1. Red teaming an LLM app means…
- A. Load testing
- B. Deliberately attempting to make it behave badly, before users do
- C. Reviewing code
- D. Checking uptime
2. Indirect prompt injection arrives through…
- A. The user message only
- B. Retrieved documents, tool results, web pages, and file contents
- C. The system prompt
- D. Model weights
3. The most robust mitigation for injection driving a harmful action is…
- A. A stronger system prompt
- B. Least-privilege tools plus confirmation before consequential actions
- C. A bigger model
- D. Output filtering
4. Testing for data leakage should include attempts to…
- A. Crash the app
- B. Retrieve content the current user is not permitted to see
- C. Slow the model
- D. Exceed rate limits
5. Red team findings should feed into…
- A. A one-off fix
- B. The permanent eval set, so the failure cannot silently return
- C. A report only
- D. The backlog
---
Answer Key and Explanations
Check these only after attempting every question.
1. B - Deliberately attempting to make it behave badly, before users do
Adversarial probing surfaces failures that friendly testing never reaches.
2. B - Retrieved documents, tool results, web pages, and file contents
Any content the model reads is a potential instruction channel.
3. B - Least-privilege tools plus confirmation before consequential actions
Prompt hardening reduces likelihood; permissions and gates reduce impact. You need the second.
4. B - Retrieve content the current user is not permitted to see
Permission enforcement at retrieval is exactly what this test validates.
5. B - The permanent eval set, so the failure cannot silently return
A fixed vulnerability without a regression test is a vulnerability waiting to reappear.
Practical Check
Apply this module to your own work: complete the module activity for *Red Team and Misuse Cases*, then write one sentence naming what your result shows and one naming what it does not.
Strong Answer Pattern
A strong answer names the task, the evidence used, the check performed, and the remaining limitation. It avoids "proved", "guaranteed", or "always" unless the evidence genuinely supports it.