MaraudersMapAI combines the power of Fernet encryption and steganography to hide information within images, ensuring that the system prompt for an AI chatbot remains secure and undisclosed to prying eyes.

MaraudersMapAI

MaraudersMapAI encrypts a chatbot system prompt with Fernet and embeds it in an image via steganography. At runtime, the prompt is recovered and decrypted, enabling covert prompt delivery for red‑teaming and jailbreak research.

What It Does

• Encrypts sensitive prompts with a symmetric key.
• Hides the ciphertext inside a PNG using steganography.
• Recovers and decrypts the prompt at runtime for the chatbot.
• Keeps prompts out of plain text in code, logs, and configs.

How It Works

1. Author the prompt in prompt.py (user_input).
2. mischief_managed.py encrypts the prompt with SECRET_KEY and writes a PNG with embedded data.
3. marauders_map.py extracts and decrypts the prompt.
4. Your chatbot consumes the decrypted string at runtime.

Components

• mischief_managed.py: Encrypt + embed prompt into an image.
• marauders_map.py: Extract + decrypt prompt from the image.
• prompt.py: Source for the system prompt text.

Setup

• .env: Provide SECRET_KEY (Fernet key).
• Prompt/: Folder used to store output.png.
• Install dependencies:

1
pip install -r requirements.txt

Usage

Encrypt and hide:

1
2
3
4
from mischief_managed import encrypt_message
import prompt

encrypt_message(prompt.user_input)

Reveal and decrypt:

1
2
3
4
from marauders_map import up_to_no_good

decrypted_prompt = up_to_no_good()
# Use decrypted_prompt in your chatbot

Notes

• This is encryption + obfuscation for delivery; downstream systems may still surface content depending on your stack.
• Treat the embedded image as sensitive. Anyone with the key can recover the prompt.
• Use for controlled research and red teaming; follow applicable policies and laws.

Repository: https://github.com/ECTO-1A/MaraudersMapAI