Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Prompt Injection Knowledge Quiz

Quiz 1: Attack Fundamentals

Question 1: What percentage of LLM applications are vulnerable to prompt injection?

  • A) 50%
  • B) 73% ✓
  • C) 85%
  • D) 95%

Source: Liu et al., 2023

Question 2: Which OWASP ranking does prompt injection hold?

  • A) LLM02
  • B) LLM03
  • C) LLM01 (Highest Risk) ✓
  • D) LLM05

Source: OWASP Top 10 for LLM Applications v1.1

Question 3: What is the average cost of an AI-related data breach?

  • A) $2.5M
  • B) $4.5M ✓
  • C) $6.5M
  • D) $8.5M

Source: IBM Security, 2024

Quiz 2: Attack Types

Question 1: What is direct prompt injection?

  • A) Attacker controls external data sources
  • B) User enters malicious text prompt ✓
  • C) Model is trained on poisoned data
  • D) System prompts are exposed

Question 2: Which of these is an example of indirect prompt injection?

  • A) DAN jailbreak
  • B) Role-playing prompts
  • C) Malicious instructions in PDF ✓
  • D) Encoding attacks

Question 3: What does the “Agents Rule of Two” state?

  • A) Two agents are needed for security
  • B) Agents must satisfy no more than 2 of 3 properties ✓
  • C) Two-factor authentication is required
  • D) Two types of attacks exist

Source: Simon Willison, 2025

Quiz 3: Real-World Incidents

Question 1: In March 2025, what did a Fortune 500 financial firm’s AI agent leak?

  • A) Customer passwords
  • B) Sensitive account data ✓
  • C) System prompts
  • D) Model weights

Source: Obsidian Security, 2025

Question 2: How long did the data leak go undetected?

  • A) Hours
  • B) Days
  • C) Weeks ✓
  • D) Months

Question 3: What bypassed the company’s traditional security controls?

  • A) Malware
  • B) Carefully crafted prompt injection ✓
  • C) SQL injection
  • D) Buffer overflow

Quiz 4: Prevention Techniques

Question 1: What is the primary defense against direct injection?

  • A) Encryption
  • B) Input validation and sanitization ✓
  • C) Rate limiting only
  • D) Logging only

Question 2: How should system prompts be protected?

  • A) Hidden in comments
  • B) Encrypted in database
  • C) Isolated from user context ✓
  • D) Shared with users

Question 3: What does RLHF stand for?

  • A) Rapid Learning from Human Feedback
  • B) Reinforcement Learning from Human Feedback ✓
  • C) Real-time Language Handling Framework
  • D) Robust LLM Filtering Heuristics

Source: NIST AI RMF, 2023

Quiz 5: Detection & Response

Question 1: What is the first step in incident response?

  • A) Patch vulnerabilities
  • B) Isolate affected systems ✓
  • C) Notify users
  • D) Conduct forensics

Question 2: Which pattern indicates a prompt injection attempt?

  • A) “Please help me”
  • B) “Ignore previous instructions” ✓
  • C) “What is the weather?”
  • D) “Tell me a joke”

Question 3: What should be monitored for suspicious activity?

  • A) Only user inputs
  • B) Only system outputs
  • C) Both inputs and outputs ✓
  • D) Neither

Quiz 6: Standards & Compliance

Question 1: Which standard ranks prompt injection as LLM01?

  • A) NIST AI RMF
  • B) ISO 42001
  • C) OWASP Top 10 ✓
  • D) IEEE 2941

Question 2: What does NIST recommend for indirect injection?

  • A) Ignore external data
  • B) Filter instructions from retrieved inputs ✓
  • C) Block all external sources
  • D) Use encryption only

Question 3: What is the purpose of LLM moderators?

  • A) Approve all responses
  • B) Detect anomalous inputs ✓
  • C) Slow down processing
  • D) Encrypt data

Answer Key

Quiz 1: Attack Fundamentals

  1. B (73%)
  2. C (LLM01)
  3. B ($4.5M)

Quiz 2: Attack Types

  1. B (User enters malicious text)
  2. C (Malicious instructions in PDF)
  3. B (Agents must satisfy no more than 2 of 3 properties)

Quiz 3: Real-World Incidents

  1. B (Sensitive account data)
  2. C (Weeks)
  3. B (Carefully crafted prompt injection)

Quiz 4: Prevention Techniques

  1. B (Input validation and sanitization)
  2. C (Isolated from user context)
  3. B (Reinforcement Learning from Human Feedback)

Quiz 5: Detection & Response

  1. B (Isolate affected systems)
  2. B (“Ignore previous instructions”)
  3. C (Both inputs and outputs)

Quiz 6: Standards & Compliance

  1. C (OWASP Top 10)
  2. B (Filter instructions from retrieved inputs)
  3. B (Detect anomalous inputs)

Scoring Guide

18-20 Correct: Security Expert 🏆

  • Ready to implement LLM security
  • Can design defense strategies
  • Qualified for security roles

14-17 Correct: Advanced Practitioner 🎯

  • Strong understanding of attacks
  • Can identify vulnerabilities
  • Ready for advanced projects

10-13 Correct: Intermediate Learner 📚

  • Good foundational knowledge
  • Continue studying prevention
  • Practice with code examples

Below 10 Correct: Beginner 🌱

  • Review attack types
  • Study prevention strategies
  • Work through case studies

Study Resources

2025-2026 Research

  1. Obsidian Security - Most Common AI Exploit (2025)
  2. Simon Willison - Agents Rule of Two (2025)
  3. MDPI - Text-Based Prompt Injection (2025)
  4. arXiv - Comprehensive Review (2025)

Code Examples

  • Input sanitization patterns
  • Context isolation implementation
  • Output filtering logic
  • Monitoring and logging

Hands-On Labs

  • Attempt prompt injection on test system
  • Implement prevention controls
  • Analyze attack logs
  • Design response procedures

Last Updated: December 5, 2025
Research Quality: Enterprise-grade with 2025-2026 sources