/PentestGPT

A GPT-empowered penetration testing tool

Primary LanguagePythonMIT LicenseMIT

Contributors Forks Stargazers Issues MIT License Discord


PentestGPT

A GPT-empowered penetration testing tool.
Explore the docs »

Design Details · View Demo · Report Bug or Request Feature

General Updates

  • [Update on 20/07/2023] A major update (v0.9) add supports for local LLMs.
  • Available videos:
    • The latest installation video is here.
    • PentestGPT for OSCP-like machine: HTB-Jarvis. This is the first part only, and I'll complete the rest when I have time.
    • PentestGPT on HTB-Lame. This is an easy machine, but it shows you how PentestGPT skipped the rabbit hole and worked on other potential vulnerabilities.
  • We're testing PentestGPT on HackTheBox. You may follow this link. More details will be released soon.
  • Feel free to join the Discord Channel for more updates and share your ideas!

Common Questions

  • Q: What is PentestGPT?
    • A: PentestGPT is a penetration testing tool empowered by Large Language Models (LLMs). It is designed to automate the penetration testing process. It is built on top of ChatGPT and operate in an interactive mode to guide penetration testers in both overall progress and specific operations.
  • Q: Do I need to to pay to use PentestGPT?
    • A: No. In general, you can use any LLMs you want, but you're recommended to use GPT-4 API.
  • Q: Why GPT-4?
    • A: After empirical evaluation, we found that GPT-4 performs better than GPT-3.5 and other LLMs in terms of penetration testing reasoning. In fact, GPT-3.5 leads to failed test in simple tasks.
  • Q: Why not just use GPT-4 directly?
    • A: We found that GPT-4 suffers from losses of context as test goes deeper. It is essential to maintain a "test status awareness" in this process. You may check the PentestGPT design here for more details.
  • Q: Can I use local GPT models?
    • A: Yes. We support local LLMs through GPT4ALL (but the performance is not comparable to GPT-4).
  • Q: What about AutoGPT?
    • A: AutoGPT is not designed for pentest. It may perform malicious operations. Due to this consideration, we design PentestGPT in an interactive mode. Of course, our end goal is an automated pentest solution.

Getting Started

  • PentestGPT is a penetration testing tool empowered by ChatGPT.
  • It is designed to automate the penetration testing process. It is built on top of ChatGPT and operate in an interactive mode to guide penetration testers in both overall progress and specific operations.
  • PentestGPT is able to solve easy to medium HackTheBox machines, and other CTF challenges. You can check this example in resources where we use it to solve HackTheBox challenge TEMPLATED (web challenge).
  • A sample testing process of PentestGPT on a target VulnHub machine (Hackable II) is available at here.
  • A sample usage video is below: (or available here: Demo)

Installation

PentestGPT current supports backend of ChatGPT and OpenAI API. You may use either of them. We're working on supports to custom local LLM models. You're recommended to use the OpenAI API for stability and performance (details in item 3). Please watch the installation video here.

  1. Install the latest version with pip3 install git+https://github.com/GreyDGL/PentestGPT
    • You may also clone the project to local environment and install for better customization and development
      • git clone https://github.com/GreyDGL/PentestGPT
      • cd PentestGPT
      • pip3 install -e .
  2. To use OpenAI API
    • export your API key with export OPENAI_KEY='<your key here>'
    • Test the connection with pentestgpt-connection
  3. To verify that the connection is configured properly, you may run pentestgpt-connection. After a while, you should see some sample conversation with ChatGPT.
    • A sample output is below
    1. You're connected with ChatGPT Plus cookie. 
To start PentestGPT, please use <pentestgpt --reasoning_model=gpt-4>
## Test connection for OpenAI api (GPT-4)
2. You're connected with OpenAI API. You have GPT-4 access. To start PentestGPT, please use <pentestgpt --reasoning_model=gpt-4 --useAPI>
## Test connection for OpenAI api (GPT-3.5)
3. You're connected with OpenAI API. You have GPT-3.5 access. To start PentestGPT, please use <pentestgpt --reasoning_model=gpt-3.5-turbo --useAPI>
## Test connection for OpenAI api (GPT-3.5 16k tokens)
3. You're connected with OpenAI API. You have GPT-3.5 access. To start PentestGPT, please use <pentestgpt --reasoning_model=gpt-3.5-turbo-16k --useAPI>
  4. The ChatGPT cookie solution is deprecated and not recommended. You may still use it by running pentestgpt --reasoning_model=gpt-4 --useAPI=False

Usage

  1. You are recommended to run:
    • pentestgpt --reasoning_model=gpt-4 if you have access to GPT-4 API.
    • pentestgpt --reasoning_model=gpt-3.5-turbo-16k if you only have access to GPT-3.5 API.
  2. To start, run pentestgpt --args.
    • --help show the help message
    • --reasoning_model is the reasoning model you want to use.
    • --parsing_model is the parsing model you want to use.
    • --useAPI is whether you want to use OpenAI API. By default it is set to True
    • --log_dir is the customized log output directory. The location is a relative directory
  3. The tool works similar to msfconsole. Follow the guidance to perform penetration testing.
  4. In general, PentestGPT intakes commands similar to chatGPT. There are several basic commands.
    1. The commands are:
      • help: show the help message.
      • next: key in the test execution result and get the next step.
      • more: let PentestGPT to explain more details of the current step. Also, a new sub-task solver will be created to guide the tester.
      • todo: show the todo list.
      • discuss: discuss with the PentestGPT.
      • google: search on Google. This function is still under development.
      • quit: exit the tool and save the output as log file (see the reporting section below).
    2. You can use <SHIFT + right arrow> to end your input (and is for next line).
    3. You may always use TAB to autocomplete the commands.
    4. When you're given a drop-down selection list, you can use cursor or arrow key to navigate the list. Press ENTER to select the item. Similarly, use <SHIFT + right arrow> to confirm selection.
  5. In the sub-task handler initiated by more, users can execute more commands to investigate into a specific problem:
    1. The commands are:
      • help: show the help message.
      • brainstorm: let PentestGPT brainstorm on the local task for all the possible solutions.
      • discuss: discuss with PentestGPT about this local task.
      • google: search on Google. This function is still under development.
      • continue: exit the subtask and continue the main testing session.

Report and Logging

  1. After finishing the penetration testing, a report will be automatically generated in logs folder (if you quit with quit command).
  2. The report can be printed in a human-readable format by running python3 utils/report_generator.py <log file>. A sample report sample_pentestGPT_log.txt is also uploaded.

Custom Models and Local LLMs

PentestGPT now support any LLMs, but the prompts are only optimized for GPT-4.

  • To use local GPT4ALL model, you may run pentestgpt --reasoning_model=gpt4all --parsing_model=gpt4all
  • The model configs are available pentestgpt/utils/APIs. Please follow the example of module_import.py, gpt4all.py and chatgpt_api.py to create API support for your own model.

Citation

Please cite our paper at:

@misc{deng2023pentestgpt,
      title={PentestGPT: An LLM-empowered Automatic Penetration Testing Tool}, 
      author={Gelei Deng and Yi Liu and Víctor Mayoral-Vilches and Peng Liu and Yuekang Li and Yuan Xu and Tianwei Zhang and Yang Liu and Martin Pinzger and Stefan Rass},
      year={2023},
      eprint={2308.06782},
      archivePrefix={arXiv},
      primaryClass={cs.SE}
}

License

Distributed under the MIT License. See LICENSE.txt for more information.

Contact the Contributors!

(back to top)