One challenge in answering the Department of Defense's call to improve the capabilities of integrated training environments is populating training scenarios with the appropriate quality and quantity of realistic entities (friendly, adversarial, and neutral). Employing entities in Live, Virtual, and Constructive (LVC) training environments is commonly accomplished using combinations of live role players and constructive entities (i.e., Semi-Automated Forces [SAFs]). This currently requires qualified personnel to role-play in complex tactical scenarios, which creates fiscal challenges (e.g., increase in manpower and material costs). A less costly solution is to increase the use of SAFs and decrease human role-player requirements; however, current SAFs possess a limited range of behaviors, lack individual differences, and require a high amount of instructor/operator interaction (e.g., pre-scripting, monitoring, scenario control). For these reasons, SAFs are only as adaptive and interactive as the instructor/operator's capabilities (i.e., workload, willingness, attention). Additionally, no standardized requirements exist for the next generation of SAFs due to a lack of common direction and goals. This lack of requirements and standards yields SAFs that ultimately lack the realism to provide effective training without excessive intervention. This paper outlines challenges of current SAF technologies, introduces a guidance model for standardizing SAF development and assessment (Human Behaving SAFs, [HBSs]) for developers and entity authors, and identifies areas of ongoing science and technology development. Ultimately, if followed, this HBS guidance model will produce future SAFs capable of enhancing the efficiency and effectiveness of LVC training, thus reducing workload associated with SAF management, and alleviating fiscal concerns.