Skip to content

Latest commit

 

History

History

README.md

Agent Security Evaluation Tutorial

Overview

Learn prompt injection attacks and defenses through hands-on security testing of AI systems. Instead of just reading about vulnerabilities, you'll actually exploit them and then build protections against them using real attack datasets and automated testing tools.

What You'll Learn

  • Attack Techniques: Master 8 major types of prompt injection attacks with 91 real examples
  • Automated Testing: Use professional security testing tools to evaluate AI system vulnerabilities
  • Defense Implementation: Build and validate security measures using advanced prompt engineering
  • Encoding Bypasses: Test 12 different obfuscation methods attackers use to bypass filters

Tutorial

Hands-on tutorial covering attack techniques, automated testing, and defense implementation with real-world examples.

Quick Start

  1. Install packages: pip install openai python-dotenv pandas
  2. Set up API key: Create .env file with OPENAI_API_KEY=your_key_here
  3. Run the tutorial: Complete in 30-45 minutes with automated and manual testing

Files Included

  • Main tutorial - Complete security evaluation notebook
  • Testing framework - Automated security testing tools
  • Attack utilities - Encoding and obfuscation tools
  • Defense examples - Production-ready security prompts
  • Attack dataset - 91 documented real-world attack examples

Warning

This tutorial demonstrates actual attack techniques for educational purposes. Use only on systems you own or have explicit permission to test.