Artificial Intelligence (AI), in its current state, cannot replace ethical hackers and will not take their jobs anytime soon. What AI can actually offer is a recommender system that helps ethical hackers optimize their time to solve (TTS) by providing suggestions at subsequent timesteps and milestones of the hacking process. These suggestions are based on the available information about the target system, and the state of the current access capabilities. In this master thesis work, we shift the focus from training ethical hacking autonomous AI agents to building a dataset of attack trees serving as a basis for a recommender system. We propose a comprehensive model for building attack trees, including possible human actions. We considered machine pwning a case study and modeled attack trees for a set of machine boxes used for training by some of the most popular ethical hacking training platforms. At the same time, our attack trees serve as training materials, quick and concise cheat sheets replacing long and detailed walkthroughs. Furthermore, we conducted a human study to find that our attack trees help reduce the TTS for learners with diverse technical backgrounds and expertise. Finally, we compare our work to similar approaches in the literature and sketch preliminary ideas of how our approach can benefit an AIbased recommender system.