nasim 0.7.3

A simple and fast simulator for remote network pen-testing

Status: Stable release. No extra development is planned, but still being maintained (bug fixes, etc).

Network Attack Simulator

Network Attack Simulator (NASim) is a simulated computer network complete with vulnerabilities, scans and exploits designed to be used as a testing environment for AI agents and planning techniques applied to network penetration testing.

Installation

The easiest way to install the latest version of NASim hosted on PyPi is via pip:

$ pip install nasim

To install dependencies for running the DQN test agent (this is needed to run the demo) run:

$ pip install nasim[dqn]

To get the latest bleeding edge version and install in development mode see the Install docs

Demo

To see NASim in action, you can run the provided demo to interact with an environment directly or see a pre-trained AI agent in action.

To run the tiny benchmark scenario demo in interactive mode run:

$ python -m nasim.demo tiny

This will then run an interactive console where the user can see the current state and choose the next action to take. The goal of the scenario is to compromise every host with a non-zero value.

See here for the full list of scenarios.

To run the tiny benchmark scenario demo using the pre-trained AI agent, first ensure the DQN dependencies are installed (see Installation section above), then run:

$ python -m nasim.demo tiny -ai

Note: Currently you can only run the AI demo for the tiny scenario.

Documentation

The documentation is available at: https://networkattacksimulator.readthedocs.io/

Using with OpenAI gym

NASim implements the Open AI Gym environment interface and so can be used with any algorithm that is developed for that interface.

See Starting NASim using OpenAI gym.

Authors

Jonathon Schwartz - [e-mail]

License

MIT © 2020, Jonathon Schwartz

What’s new

• 2020-09-23 (v 0.7.3) (MICRO release)

  • Fixed issue with scenario YAML files not being included with PyPi package

  • Added final policy visualisation option to DQN and Q-Learning agents

• 2020-09-20 (v 0.7.2) (MICRO release)

  • Fixed bug with ‘re-registering’ Gym environments when reloading modules

  • Added example implementations of Tabular Q-Learning: agents/ql_agent.py and agents/ql_replay.py

  • Added Agents section to docs, along with other minor doc updates

• 2020-09-20 (v 0.7.1) (MICRO release)

  • Added some scripts for running random benchmarks and describing benchmark scenarios

  • Added some more docs (including for creating custom scenarios) and updated other docs

• 2020-09-20 (v 0.7.0) (MINOR release)

  • Implemented host based firewalls

  • Added priviledge escalation

  • Added a demo script, including a pre-trained agent for the ‘tiny’ scenario

  • Fix to upper bound calculation (factored in reward for discovering a host)

• 2020-08-02 (v 0.6.0) (MINOR release)

  • Implemented compatibility with gym.make()

  • Updated docs for loading and interactive with NASimEnv

  • Added extra functions to nasim.scenarios to make it easier to load scenarios seperately to a NASimEnv

  • Fixed bug to do with class attributes and creating different scenarios in same python session

  • Fixed up bruteforce agent and tests

• 2020-07-31 (v 0.5.0) (MINOR release)

  • First official release on PyPi

  • Cleaned up dependencies, setup.py, etc and some small fixes