The advent of large language models suggests rethinking how Agent-Based Models (ABMs) are designed, making it possible to simulate human social behavior in addition to decision making. The simulation performed by the new generative ABMs is always challenging, requiring a large number of human validators to monitor the artificial agents and check if they exhibit believable behavior. The aim of this project is to design and develop new approaches, based on natural language processing, rule-based methods, and machine learning, to reduce or remove the need for human validators for humanized ABMs validation.