While artificial intelligence (AI) offers promising approaches for developing intelligent autonomous driving (AD) agents, ensuring the safety of these AI-driven AD systems is a critical challenge. This paper proposes an approach to enhancing AD safety through the development of a safety shield based on online model checking. The safety shield acts as a real-time verification layer, monitoring and validating the actions proposed by the AI agent before execution. We demonstrate the practicality and efficiency of our approach through a highway driving case study with different AI agents trained. We construct a formal model of the driving environment, capturing the states and behaviors of the ego vehicle and surrounding traffic, and specify the safety requirements within this model. For each proposed action, we leverage Maude’s invariant model checker to determine if executing the action would violate the safety requirements. Our experimental results demonstrate the capability of online model checking to enhance the safety of AI-driven autonomous vehicles.