PromptCrypt is proposed as an interesting technique to secure LLMs in a recent paper. They mention the use of emojis, which is a great way to encrypt prompts from humans.
This Python script demonstrates how to integrate with the OpenAI API using a simple "encryption" and "decryption" mechanism for the prompts. It uses Base64 encoding as a simulated form of encryption to showcase how data might be manipulated before being sent to the API. This example specifically uses GPT-3.5, but feel free to switch between models.
- Python 3.x
requestslibrary
Ensure you have the necessary Python version and the requests library installed. You can install requests using pip if you haven't already:
pip install requests
Encryption/Decryption: The script simulates encryption and decryption using Base64 encoding and decoding, respectively.
Sending Encrypted Prompts: Encrypted prompts are decrypted back to plain text within the script before being sent to the OpenAI API.
API Request: A POST request is made to the OpenAI API with the decrypted prompt, and the response is processed and printed.
To use this script, you simply need to call the full_workflow function with a prompt as the argument