Cipher Whisperer

Overview

Cipher Whisperer is a software project that aims to decrypt and analyze cipher texts using a graph-based approach. The project utilizes various software libraries and tools to process and transform the input text.

Graph Architecture:

The system takes in a text, extracts a single cipher text block from it and detects what suspected ciphers it matches with a probability of above 70% confidence level. Lastly the system calls the respective decryptors on the cipher text in an attempt to decrypt it correctly.

Software and Libraries Used

• Python 3 as the primary programming language
• dotenv for environment variable management
• langgraph for graph-based processing and transformation
• langchain_core and langchain_openai for natural language processing and AI-powered decryption
• pydantic for data validation and parsing

Decryptors

The project features a range of decryptors, including:

• BruteForceDecryptor: a brute-force decryptor that uses a list of English words to decrypt the follow ciphers: Caesar, Atbash, Affine, Bacon, Transposition
• Other decryptors (to be added in future updates)

Sample run:

Future work include:

• Adding a reflection component to the system that analyzes the output after the first run and tries again.
• Adding the ability to extract multiple cipher text blocks from a given text.

Getting Started

Prerequisites

• Python 3
• Required libraries: dotenv, langgraph, langchain_core, langchain_openai, pydantic

Installation

1. Clone this repository.
2. Create a virtual environment.
3. Install the required libraries: pip install -r requirements.txt
4. Create a .env file with your environment variables (i.e. API keys) following the .env.example file

Running the Code

1. Run the main script: python main.py
2. Follow the prompts to input a cipher text and receive the decrypted output

Project Structure

• cipherwhisperer/: main project directory
  • main.py: entry point for the project
  • graph.py: graph-based processing and transformation of input text
  • chains.py: chain-based processing and transformation of input text
  • models.py: data models for the project (e.g. GraphState)
  • decryptors.py: decryptor implementations (e.g. BruteForceDecryptor)
  • .gitignore: ignore file for Git

Contributing

Contributions are welcome! Please submit a pull request with your changes and a brief description of what you've added or fixed.

License

This project is licensed under the MIT License. See LICENSE for details.

Reference

1. Yuan, Y., Jiao, W., Wang, W., Huang, J., He, P., Shi, S., & Tu, Z. (2024). GPT-4 Is Too Smart To Be Safe: Stealthy Chat with LLMs via Cipher. In The Twelfth International Conference on Learning Representations Link Here.