Skip to content

NetSecGame

The NetSecGame (Network Security Game) is a framework for training and evaluation of AI agents in network security tasks. It supports both offensive and defensive operations in a dynamic, multi-agent environment built on top of the CYST network simulator.

Key Features

  • Multi-agent support — Multiple attackers, defenders, and benign users can interact simultaneously in real-time.
  • Configurable scenarios — Choose from predefined network topologies or define custom ones using CYST configurations.
  • Standard RL interface — Agents submit Actions and receive Observations with state, reward, and terminal flag.
  • Rich game state — The GameState captures networks, hosts, services, data, and firewall blocks — far richer than flat vector representations.
  • Stochastic global defender — A built-in SIEM-like defender provides realistic opposition without requiring a trained agent.
  • Dynamic topologies — Optionally randomize IP addresses between episodes to prevent overfitting.
  • Trajectory recording — Record and analyze full episode trajectories for debugging and research.
Getting Started Installation, configuration, and running your first game
Architecture Game components, actions, preconditions, and observations
Configuration Detailed environment and task configuration reference
Global Defender Stochastic detection system and thresholds
API Reference Auto-generated API documentation
NetSecGameAgents Reference agent implementations (Random, Tabular, LLM, DQN)
GitHub Project source code, issue tracker, and contributions
PyPI Latest NetSecGame releases on the Python Package Index

About

This project was developed at the Stratosphere Laboratory at the Czech Technical University in Prague. The project is supported by Strategic Support for the Development of Security Research in the Czech Republic 2019–2025 (IMPAKT 1) program, by the Ministry of the Interior of the Czech Republic under No. VJ02010020 – AI-Dojo: Multi-agent testbed for the research and testing of AI-driven cyber security technologies.