Users’ mental models of security, though possibly incorrect, embody patterns of reasoning about security that lead to systematic behaviors across tasks and may be shared across populations of users. Researchers have identified widely held mental models of security, usually with the purpose of improving communications and warnings about vulnerabilities. Here, we implement previously identified models in order to explore their use for predicting user behavior. We describe a general approach for implementing the models in agents that simulate human behavior within a network security test bed, and show that the implementations produce behaviors similar to those of users who hold them. The approach is relatively simple for researchers to implement new models within the agent platform to experiment with their effects in a multi-agent setting.