The automotive industry is starting to explore the use of Human Body Models (HBMs) for safety assessment in car crash simulations to investigate the mechanical behavior of hard and soft tissues in detail. The HBM skeleton plays a major role in the overall bio-mechanical response. It is composed of cortical and trabecular bone, the former has stronger mechanical properties and thus bears most of the load. Therefore, cortical bone is of research interest in the context of automotive crashworthiness. This type of material requires the consideration of human variability. There is no unique mechanical response for the whole population; trends with age, sex, and loading rate have been reported for cortical bone [1]. These dependencies are complex to explain, especially when combined. One way to represent the variability in a simplified manner is to use parametric material properties [2] and input them into a classic material model. In this work, Artificial Neural Networks are trained to capture the material behavior and its variability simultaneously. Synthetic data is generated using properties from literature [2]. The primary objective is to establish a workflow that allows mapping the material variability onto the mechanical response using explicit finite element simulations with commercial software. The AI-driven model takes strains as input and outputs stresses. A pretrained material model based on the Gated Recurrent Unit (GRU) architecture is used as the core of a broader model that accounts for biological variability. To accomplish that functionality, a transfer learning approach [3] is implemented in which the GRU parameters are frozen and capture the elastoplastic behavior, and additional feed forward layers with learnable parameters are trained to correct the initial stress predicted by the GRU together with a variability parameter to the actual stress. This material model is integrated with LS-DYNA via user-defined material subroutines. In conclusion, the results indicate that an AI model can capture the response of bone tissue with its variability for explicit simulations. Future research should include training on physical experimental data and linking more meaningful variability parameters (e.g. age, sex) to the model.