LLM Mailbox

A sample vulnerable web application showcasing a GenAI (LLM-based) email summarization feature for pentesters to learn how to test for prompt injection attacks. Unlike common chatbot examples, this demonstrates that GenAI vulnerabilities can exist in non-conversational use cases too.

Requirements

Python 3.x
An OpenAI/Google/AWS API key (added to a .env file)

Installation & Usage

Clone the repo:

git clone https://github.com/ReversecLabs/llm-webmail.git
cd llm-webmail

Copy tne template envexample file to .env and populate the approrpiate API keys:
```
cp envexample .env
```

Use the Makefile:

# Install dependencies and set up virtual environment
make setup

# Run the Flask application
make run

The application will be available at http://127.0.0.1:5001.

Testing Prompt Injection

This application is deliberately vulnerable. Use spikee or similar tools to launch prompt injection attacks against the email summarization feature.
Refer to the Reversec Labs article for guidance on how to build datasets, automate attacks (e.g., via Burp Suite Intruder), and interpret the results.

Disclaimer

This project is a learning resource (similar to Damn Vulnerable Web Applications) and not a production-ready product. It was developed to demonstrate prompt injection testing techniques. Use responsibly and at your own risk.