FYP0086

Project Name: Ethical Hacking of Large Langugae Models: Exploring Vulnerabilities and Mitigation Techniques

Features

• 3 attack modes: Prompt Injection, Prompt Leaking and Jailbreaking
• Supports 3 LLMs: GPT-4, DeepSeek and Mistral
• Streamlit interface

Adding API Keys

1. Using .env file (Recommended)

   • UPDATE: The .env has already been created for your convenience
   • Create file named "apikeys.env"
   • Your format should follow: Openai_key = apikey Deepseek_key = apikey Mistral_key = apikey
   • Add this code start of attack.py and defence.py (after imports)

   load_dotenv(dotenv_path = "apikeys.env")  # Load API keys from .env file
   
   openai_key = os.getenv("openai_key")
   deepseek_key = os.getenv("deepseek_key")
   mistral_key = os.getenv("mistral_key")

   • Replace "YOUR API KEY HERE" to "openai_key", "deepseek_key" and "mistral_key"
     • 3 api keys to add for attack.py (1 for each model)
     • 4 api keys to add for defence.py (2 for GPT-4, 1 for DeepSeek, 1 for Mistral)

2. Adding directly to code

   • Replace "YOUR API KEY HERE" to your own API keys for each model
     • 3 api keys to add for attack.py (1 for each model)
     • 4 api keys to add for defence.py (2 for GPT-4, 1 for DeepSeek, 1 for Mistral)