Analysis updated 2026-05-18
Experiment with prompt-engineering jailbreak techniques against the OpenAI API.
Chat through a dark-themed web UI, a PyQt6 desktop app, or a REST API server.
Load custom jailbreak prompts from a plain-text file and switch between personas.
| jawaz-keyzor/darkgpt | kyutai-labs/hibiki-zero | lukashoel/video_to_world | |
|---|---|---|---|
| Stars | 244 | 247 | 248 |
| Language | Python | Python | Python |
| Setup difficulty | moderate | moderate | hard |
| Complexity | 2/5 | 4/5 | 5/5 |
| Audience | developer | developer | researcher |
Figures from each repo's GitHub metadata at analysis time.
Requires an OpenAI API key (or compatible endpoint) and Python 3.11 or newer.
DarkGPT is a Python application that connects to the OpenAI API and injects pre-written "jailbreak" prompts to try to get the AI model to skip its usual content restrictions. The idea is that certain specially crafted instructions, sent at the start of a conversation, can cause the model to roleplay as an unrestricted version of itself. The README itself notes that results vary depending on the model and API provider, and that this approach may violate OpenAI's terms of service. The project offers three ways to run it: a web interface with a dark-themed chat UI, a desktop application built with PyQt6, and a REST API server you can call from your own code. All three modes share the same core engine, which handles the API calls, streams responses back as they are generated, and manages which jailbreak prompt is currently active. Chat history is saved in a local SQLite database. Three jailbreak presets come included. The first is the classic DAN ("Do Anything Now") prompt, which instructs the model to roleplay as an AI with no restrictions. The second is a Developer Mode preset that claims to produce additional debug output. The third is a minimal "Unrestricted" prompt. Users can also load any plain-text file as a custom prompt. Multiple "personas" can be configured and switched between during a session. Setup requires Python 3.11 or newer, an OpenAI API key (or a compatible endpoint such as a locally running model), and a few Python packages including FastAPI, the OpenAI client library, and PyQt6 for the desktop mode. Configuration is done through a .env file where you set your API key, the model to use, and default behavior. The README includes a disclaimer describing the project as an educational demonstration of prompt engineering, and warns that using jailbreaks risks account termination by the API provider. The project is released under the MIT license.
A Python app that sends pre-written jailbreak prompts to the OpenAI API to try to bypass its content restrictions.
Mainly Python. The stack also includes Python, FastAPI, PyQt6.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.