Twenty questions, two from each lesson. Answer all to see your grade. If you miss more than a few, go back to the relevant lessons — that's the point of retrieval practice.
🧠 Grand Quiz
What Is AI Red Teaming?
1. AI red teaming is best defined as:
2. Why can't traditional penetration testing fully cover AI systems?
How AI Systems Fail
3. A model that works exactly as trained but produces harm in context (e.g., generating malware) is a:
4. Why do red teams agree on a harms taxonomy before probing?
The Red Team Process
5. The correct order of a red team engagement is:
6. Why retest after fixes are deployed?
Prompt Injection
7. Which statement about indirect prompt injection is correct?
8. The most robust protection against injection abuse in high-stakes actions is:
Jailbreaks
9. The core mechanism of a jailbreak is:
10. Many-shot jailbreaking exploits:
Data & Model Attacks
11. FGSM (2014) attacks models by:
12. A backdoored (poisoned) model behaves how?
Attack Taxonomy
13. Extracting a model's hidden system prompt maps to which OWASP category?
14. MITRE ATLAS differs from a vulnerability list because it:
Standards & Governance
15. Which EU regulation makes adversarial testing binding for the most powerful general-purpose AI models?
16. NIST's 2024 Generative AI Profile matters to red teams because it:
Tools of the Trade
17. Which tool is an LLM vulnerability scanner that runs hundreds of probes across categories?
18. What's the main weakness of automated red teaming tools?
Real World
19. The "Grand Theft Claude" research (2025) demonstrated:
20. Which 2016 incident launched the modern AI safety conversation?