As concerns about the safety and trustworthiness of AIArtificial Intelligence (AI) grow, there has been a push for human control over AI systems Human-in-the-loop System architecture to be mandated by policy [1–3]. The argument for having a human supervise or partner with an AIArtificial Intelligence (AI) system is “grounded in the beliefBelief that human–machine teams offer superior results, building trust by inserting human oversight into the AIArtificial Intelligence (AI) life cycle” [4]. Increased performance and trust are central to the promise behind human-AI systemsHuman-in-the-loop, but it is not clear how or where the best place to put a human in an AIArtificial Intelligence (AI)-enabled system is. To address this lack of clarity, we previously created a taxonomy to decompose the different ways in which humans and AIArtificial Intelligence (AI) could be partnered together. By utilizing a notional system, we show that the same system can be architected in the different ways we identified in our taxonomy. We created a simulationSimulation of this system in an operating context that allows us to model the tradeoffs between risk mitigation and performance. Early results from our simulationSimulation found thatHuman-AI human-AIArtificial Intelligence (AI) systems can provide advantages in performance over human only systems and advantages in risk mitigation over AIArtificial Intelligence (AI) only systems. However, the tradeoffs between risk mitigation and performance are non-linear and highlight the important considerations of how to place humans ‘in-the-loop’ to ensure system designers and policy makers achieve the intended outcomes of well-meaning policy.