CybORGView: An Interactive Interface for Visualizing Reinforcement Learning Agent Performance in Autonomous Cyber Operations

Abstract

Autonomous Cyber Operations (ACO) increasingly leverage Reinforcement Learning (RL) to train agents capable of making effective decisions, where success is measured through a scalar reward signal. However, reliance on rewards alone obscure agent behavior, which directly hinders development efficiency and reduces confidence in operational deployment. In this paper, we present CybORGView, the first visualization tool for ACO to our knowledge that provides full visibility to an agent's performance beyond abstract reward signals. CybORGView allows ACO practitioners to analyze action distributions, red activity, and policy convergence across entire training iterations in a concise, graphical format. Through our intuitive interface, developers can easily tune ACO agents, lowering the barrier of technical expertise required. This visibility facilitates more reliable agent development and evaluation, advancing the path toward practical ACO deployment.

Keywords: Autonomous Cyber Operations, Autonomous Cyber Defence, Reinforcement Learning, Cybersecurity, Visualization, Artificial Intelligence

DOI: 10.54941/ahfe1007178