Conference:
4th IEEE Conference on Information Technology and Data Science (CITDS 2026) 27–28. August 2026, Debrecen, Hungary
Authors:
Szilvágyi D, Buttyán L.
Abstract:
Attack trees support threat modeling by decomposing high-level attacker goals into smaller subgoals connected by AND/OR logic. Manual construction is time-consuming and does not scale to complex systems. This paper presents a rulebased generator that builds attack trees automatically from a system model represented as typed entities and their relations. Starting from a goal predicate, the generator performs backward chaining over precondition rules, producing an AND/OR dependency structure whose leaves are either model facts or basic attack steps that represent concrete attacker actions. To avoid combinatorial blow-ups, the generator reuses identical substructures via memoization, turning the output into a compact directed graph. Instead of pruning recursive dependencies during generation, cycles are recorded explicitly and may become the subject of a later analysis phase. We demonstrate the proposed solution on an industrial control system and evaluate the effect of the main scalability optimizations.
