Answers are grounded in this paper and cite their sources.
Practice-oriented cybersecurity education often requires novice learners to build target web systems by hand before security analysis begins, imposing extraneous cognitive load. An earlier study introduced a Red/Blue teaming model in which a large language model (LLM) agent builds the target systems (vibe coding) under Human-on-the-Loop (HOTL) control, so that learners audit, attack, and defend. This single-cohort study examines learner-level evidence from the 2026 cohort (n = 34) of a university web-security course, set against institutional Course-Quality-Improvement (CQI) satisfaction records for nine of the ten academic years 2017–2026. Evidence comprised self-reported coding-effort logs (n = 25), single-item final-course survey measures (n = 27), and peer cross-evaluation scores (71 ratings). Logged total coding effort had a median of 3.8 h. In the final-course survey, 96% agreed implementation burden had decreased and 93% that deeper engagement in vulnerability analysis and defense design was possible (raw means 4.67 and 4.44 of 5; Wilcoxon p < 0.001). The 2026 CQI score (91.55) ranked second among the nine records (descriptive). In this single-institution context, delegating system construction to an LLM agent was associated with lower perceived implementation burden, greater reported security-analysis engagement, and high satisfaction; multi-site replication is required.
No takes yet. Share an insight, caveat, or question.
D K Kim (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: