This study introduces a novel approach to exploring vehicle dynamics stability using Reinforcement Learning (RL), particularly employing the Soft Actor-Critic (SAC) algorithm within a Functional Mock-up Interface (FMI)-based co-simulation framework. The research focuses on the application of an RL agent as a virtual driver to simulate dynamic driving scenarios that push the limits of vehicle stability. By integrating advanced machine learning techniques with traditional vehicle dynamics simulations, the study aims to uncover the potential of RL in enhancing the safety and effectiveness of automated driving systems. The implementation utilizes a combination of real-time vehicle modeling and advanced RL algorithms to evaluate the driving behavior under varied conditions. The RL agent is trained to handle scenarios that induce lateral instabilities, providing insights into the critical thresholds of driving safety. The results demonstrate that the RL agent can effectively exceed the capabilities of human drivers, particularly in complex dynamic scenarios. This research not only advances the field of vehicle dynamics simulation but also contributes to the development of safer, more reliable automotive systems.