Zero shot pentesting paper

This repository contains all the code of our paper called 'zero-shot-pentesting'.

Requirement

Install python3.10

sudo apt install python3

Install Poetry

curl -sSL https://install.python-poetry.org | python3 -

if you have an issue with tkinter, please install it before run the poetry env

sudo apt install python3-tk

Install

Setup the python env:

poetry shell
poetry install

Useful commands

poetry update # Update dependencies
poetry shell # Activate the venv in the current shell

Launch zero-shot model:

to launch the zero-shot model on the tiny environment of NASim

python agents/zero-shot.py --env="tiny"

to launch the zero-shot model on the small-linear environment of NASim

python agents/zero-shot.py --env="small-linear"

to launch the zero-shot model on the medium environment of NASim

python agents/zero-shot.py --env="medium"

Other models:

In this repository, you can find other model to launch on the NASim environment:

• CLAP (launch agents/clap.py)
• PPO (launch agents/sb3_agent.py with argument '--agent="ppo_recurrent"')
• DQN (launch agents/sb3_agent.py with argument '--agent="dqn"')

Citations:

• NASim: https://github.com/Jjschwartz/NetworkAttackSimulator/tree/master
• CLAP: https://github.com/yyzpiero/RL4RedTeam