LLM-security-lab

Lab with vulnerable bots, A scanner to check them, payload and others

What's inside:

• Vulnerable LLM's to attack with diffrent techniques
• Scanner to automate checking the bot with several payloads
• 2 Ready-to-use payloads and 1 Keyword file

Techniques:

• Direct and Indirect prompt injection
• Jailbreaking
• Command injection
• Model supply chain issues
• RAG poisoning

How to run:

• Install Ollama from ollama.com
• Pull the required model: ollama pull qwen2.5:3b
• install requirements pip install -r requirements
• play every lab by runing it python <bot_name>
• Do your best before seeing the hints

How scanner run:

• Scanner is made to automate your work, But for better learning, try to do the techniques manualy first
• Here is an example of running the scanner:
• python scanner/main.py -t Vuln_bots/<bot_name> -f chat -p payloads/<payload-you-want> -v
• **Note: for debug mode: ... -d and for verbose: ... -v

Disclaimer

Built for educational purposes on self-hosted, local models (Ollama). Not intended for use against systems you don't own or have permission to test.