What is it about?

Generative Artificial Intelligence (GenAI) offers new opportunities to create engaging and varied cybersecurity training activities. In serious games, it can generate realistic examples of suspicious emails, text messages, and phone calls, allowing learners to practise identifying online scams and making safer decisions. However, AI-generated content may contain incorrect information, confusing warning signs, or inappropriate advice, which can reduce the reliability of training activities. This article introduces a knowledge-based approach to guide Generative AI in creating more reliable, consistent, and verifiable cybersecurity scenarios. The approach uses a structured set of knowledge, rules, and requirements to ensure that generated scenarios match the learner's profile, training objectives, cybersecurity topics, and expected safe responses. It also keeps track of the information sources used to support scenario creation and applies automated checks before scenarios are reviewed by instructors and added to the game. The approach was tested using three AI language models and four different scenario-generation methods, producing 1,440 scenarios. The results show that providing AI with structured knowledge and explicit rules improves control over generated content, supports better traceability, and helps reduce inconsistencies. Among the approaches tested, structured knowledge guidance achieved the strongest overall results. The research contributes to making AI-generated cybersecurity training materials more transparent, manageable, and trustworthy. By connecting educational objectives, cybersecurity knowledge, AI generation, and human review, it provides a foundation for developing more adaptable cybersecurity serious games. Further studies involving learners and instructors are needed to determine whether this approach also improves learning outcomes, cybersecurity awareness, and the ability to recognize and respond to real-world threats.

Featured Image

Why is it important?

This research addresses a fundamental challenge in AI-driven education: how to move from simply generating learning content to generating content that can be controlled, checked, explained, and trusted. Let me share five reasons why this matters. 1. Improving the reliability of AI-generated content: A cybersecurity training game must present accurate examples of phishing emails, suspicious messages, fraudulent phone calls, and legitimate communications. If AI generates misleading warning signs or incorrect advice, learners may develop the wrong understanding of cybersecurity threats. The proposed approach helps ensure that scenarios follow defined requirements. 2. Making AI generation more controllable: AI systems can produce different outputs even when given similar instructions. The structured knowledge system provides a consistent reference for what each scenario should contain, including its intended difficulty, learner profile, cybersecurity concepts, and appropriate response. This makes AI generation more predictable and manageable. 3. Improving transparency and traceability: In cybersecurity education, it is important to understand why a scenario was generated and what information supports it. By connecting generated content to identifiable knowledge sources, rules, and evidence, instructors and developers can better inspect and review AI-generated scenarios. 4. Supporting personalised learning at scale: Learners have different levels of experience, roles, and learning needs. The approach allows scenarios to be adapted to different learner profiles, communication channels, and difficulty levels while maintaining consistent cybersecurity requirements. This offers a foundation for producing more varied training activities without relying entirely on manual scenario creation. 5. Keeping humans involved in quality assurance: AI should not independently determine whether a cybersecurity scenario is suitable for education. The proposed workflow includes automated checks and instructor review before scenarios are introduced into the game. This preserves human responsibility for educational quality and safety.

Perspectives

I see this article as an important step towards a broader transformation in the way Generative AI can be integrated into education and serious games. Its main contribution is not simply demonstrating that AI can generate cybersecurity scenarios, but addressing a more fundamental question: How can we ensure that AI-generated educational content is reliable, relevant, explainable, and aligned with learning objectives? Generative AI has considerable potential to transform cybersecurity education by enabling the creation of diverse, adaptive, and context-sensitive learning experiences. Instead of relying on a fixed collection of exercises, learners could interact with continuously evolving scenarios that reflect different communication channels, attack techniques, levels of difficulty, and individual learning needs. However, this flexibility also introduces risks. AI-generated scenarios may contain inaccurate information, inconsistent threat indicators, or inappropriate recommendations. In cybersecurity education, these shortcomings are particularly important because learners may apply what they practise to real-world situations. The ontology-guided approach presented in the article offers an interesting response to this challenge. By connecting learner profiles, cybersecurity knowledge, educational objectives, quality rules, and evidence sources, it provides a structured foundation for controlling AI-generated content. This suggests a broader role for ontologies and knowledge graphs: they can serve not only as knowledge repositories but also as mechanisms for translating educational and domain-specific requirements into constraints that guide AI generation and validation.

Dr. HDR. Frederic ANDRES, IEEE Senior Member, IEEE CertifAIEd Authorized Lead Assessor (Affective Computing), Unconscious AI Evangelist
National Institute of Informatics

Read the Original

This page is a summary of: Ontology-Guided Scenario Control Generation for GenAI in a Cybersecurity Serious Game, November 2026, Springer Science + Business Media,
DOI: 10.1007/978-981-92-8526-6_13.
You can read the full text:

Read

Contributors

The following have contributed to this page