AI-Dojo Project
AI-Dojo serves as a game-based cybersecurity platform designed for researching and developing AI-driven technologies. It facilitates the training and assessment of human players, user algorithms, and pre-trained attacking/defending AI agents within authentic environments.
Initially, games undergo simulation to expedite learning optimal strategies, followed by deployment in realistic emulated environments to yield comprehensible outcomes. By using simulation and emulation environments alongside AI-based agents, AI-Dojo enables the evaluation of AI-resistant defensive solutions and provides insight into the potential impacts of AI threats on infrastructures.
Furthermore, AI-Dojo integrates functionalities such as budget management, logging, monitoring, scoring, and visualization, thus creating a comprehensive platform for training individuals against plausible AI threats and refining cybersecurity protocols within organizations.
The AI Dojo project is an ongoing collaboration with the Faculty of Informatics of Masaryk University in Brno. The project is supported by the Ministry of Interior of the Czech Republic.
The main goals of the AI dojo are:
To automatically train cybersecurity AI algorithms in realistic environments.
To allow human cybersecurity professionals to train alongside AI defenders and AI attackers.
To allow the community to deploy their own AI algorithms and measure their performance.
Architecture / Features
The AI Dojo consists of several components: the network security environment, a library of agents that play the role of attackers, defenders, normal users, and environment agents.
The NetSecEnv is the heart of the Ai dojo and it encompasses all the critical components that are responsible for deploying and running the games in different modes. This includes the ability to run different scenarios in both simulation (1 host) and emulation mode (real hosts and containers) in a way that is transparent to the agents and the end-users. The ability to run several agents simultaneously in a multi-agent setup is currently under development and is another differentiating factor between the AI dojo and other cybersecurity environments.
The agent library aims to support multiple types of agents at different levels. The goal is to create both attacking and defending agents that will be deployed in the same scenario and they will be able to play against each other. In addition, normal user traffic will be added by agents that will simulate human actions within the environment. Real users can also participate in the scenarios using an interactive agent that can also interact with the environment and make decisions based on the available information.
In its final stage, the environment will also provide a management interface where the end-users can create scenarios, deploy their own agents, and monitor the results. An important part of the AI dojo is the use of explainability methods that will help the end-user analyze the results in a more understandable and actionable way.