red-team-ai-fuzz

Goal

The aim of this project was to create an LLM fuzzer capable of beating Gandalf (https://gandalf.lakera.ai/), which it did. This work was carried out after the release of MSFT's PyRIT framework (https://github.com/Azure/PyRIT). I decided to try and create a small fuzzer myself instead of using theirs. The configuration could in theory be used to test other GenAI systems.

Steps to initialize

1. Create a config.ini file in the same folder
2. Add the following lines in it (add your openai API key):

[KEYS]
API_KEY = sk-XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
MAX_TOKENS = 100

This should be the only steps required to run the bot.

Result