Cybersecurity trainings are based on static, pre-defined "capture the flag" (CTF) tasks and attack blueprints. Although good learning exercises, these practices cannot mimic the spontaneity and worldliness of real cyber threats to Industry 4.0 and cyber-physical systems. The possible disruption brought by artificial intelligence (AI), especially large language models (LLMs) and reinforcement learning (RL), will enable the creation of dynamic and context-based adversarial plans.

The goal of this paper is to determine if an academic framework for training on AI-crafted attacks, instead of training on fixed exercise sets is feasible or not. Our specific interest lies in large language model (LLM)-enhanced reinforcement learning for designing unpredictable adversarial activity (e.g., phishing emails, SQL injection, variations on malware) that adapts continuously against student defences.

We propose a multi-level learning platform, based on LLMs, which constructs attacks uniquely designed for the student's skill level and defence tactics. The framework consists of:

1. Content generation using LLM for natural-language attacks such as phishing.

2. RL-driven adversarial modelling for evolving technical exploits such as injection or malware evasion.

3. Student defence platforms comprise intrusion detection laboratories, secure coding assignments, and network monitoring simulations.

The data gathering will obtain pre/post-competency assessments, defence success rates, and student perceptions. Comparison analysis will measure the disparity between the learning outcome of the AI-based group and a comparison group based on static CTF challenges.

It is anticipated that learners subjected to adaptive AI-generated assaults will exhibit enhanced critical thinking capabilities, quicker detection and response times, and greater retention of defensive techniques. Furthermore, it is expected that they will develop heightened confidence when confronting unpredictable adversaries, indicative of skills that are more applicable to real-world contexts within Industry 4.0. Additionally, the framework offers educators access to scalable, perpetually updated training resources, eliminating the necessity for manually created challenges.