<p>Educators increasingly require training to address intensive challenging behavior, such as self-injury or aggression, exhibited by students with disabilities. Although ideally acquired through hands-on experience, opportunities for training in evidence-based techniques under expert supervisors are limited. Virtual reality (VR) simulations may allow educators to train without risking harm to students. However, VR typically requires personnel to assess trainees and manage the simulation, which only partially assuages concerns regarding access to training. Artificial intelligence (AI) may be integrated into instructional VR simulations to automate content generation and assessment. Yet practices used to address severe challenging behaviors, such as functional communication training (FCT), involve complex skills. This study evaluated an automated VR simulation designed to teach steps in FCT using a waitlisted randomized controlled trial. Teacher candidates (<i>n</i> = 49) were randomly assigned to an intervention (i.e., lecture, modeling, and VR) or a control group (i.e., lecture and modeling only). Trainees completed pre-, post-, and maintenance test probes during three one-to-one sessions over three weeks. Results of a mixed ANOVA indicated trainees in the VR condition exhibited more correct FCT steps at post-test (<i>d</i> = 2.45) and maintenance (<i>d</i> = 1.58) relative to the control group. Following delayed access to the VR simulation, control group participants also demonstrated a large within-subjects effect (<i>d</i> = 2.37). Surveys of knowledge and confidence did not favor either group, however. Participants reported positive perceptions of the experience overall. Results have implications for VR development and implementation in higher education.</p>