The goal of this thesis is to design, implement, and evaluate a set of three practical training modules focused on adversarial AI and its countermeasures in cybersecurity education. 

The theoretical part of the work will review the current landscape of generative AI threats, analyze the underlying technologies of selected open-source models, and compare modern AI-assisted detection mechanisms against legacy heuristic defenses. 

In the practical part, the author will (1) design structured laboratory workflows utilizing an "offensive-defensive" pedagogical paradigm covering deepfake manipulation, LLM-based social engineering, and prompt injection, (2) engineer robust and reproducible execution environments utilizing containerization in Podman, and (3) integrate state-of-the-art open-source generation models (e.g., Mistral-7B, FaceFusion) and detection algorithms (DistilBERT, LIPINC) into the exercises.  
 
Finally, the technical feasibility, architectural challenges, and educational value of the created modules will be evaluated during a pilot test with several students. Based on their feedback, the author will incorporate necessary revisions into the final output. 
 
The practical outputs of the project (source code, exercise notebooks, presentations, and infrastructure configuration files) will be published via IS MUNI under an open-source license (GNU General Public License when necessitated by third-party dependencies; MIT or CC-BY License when possible otherwise).