Year: 2026
Type: preprint
Study design: Systematic Review
Abstract: Developing realistic scenarios for Cyber Training Ranges (CTRs) requires current threat information, suitable infrastructure, attack events, learning objectives, and consideration of trainee capability. Established scenario-generation approaches use manual design, simulation, templates, rules, attack graphs, and model-driven engineering, often requiring substantial expert effort to develop and update. Recent research has investigated Artificial Intelligence (AI) for scenario generation and training, including Machine Learning (ML), Reinforcement Learning (RL), and Large Language Models (LLMs). However, existing surveys provide fragmented coverage of AI across the scenario-generation lifecycle and have not systematically examined AI-specific threat intelligence, such as MITRE ATLAS, as a dedicated source for CTR scenarios. This paper presents a Systematic Literature Review (SLR) of embedding AI into CTR scenario generation, tracing the progression from non-AI approaches to ML, LLM-based, and agentic AI methods. It critically compares their advantages and limitations and examines Retrieval-Augmented Generation (RAG), agentic approaches, and AI threat intelligence-based attack scenarios. The literature indicates that AI can support scenario creation, automation, adaptation, and deployment, while important challenges remain in threat-data quality and currency, hallucinations, technical validation, reproducibility, interoperability, and alignment with training objectives. The review identifies these limitations and discusses research directions for more realistic, adaptable, and technically reliable cybersecurity training scenarios.
Source: Research Square
Authors: Purvi Mehta, Paul Black, Gour Karmakar, Paul Pang
Institution: Federation University
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7226001544
