Mission-critical systems, such as autonomous vehicles, operate in dynamic environments where unexpected events should be managed while guaranteeing safe behavior. Ensuring the safety of these complex systems is a major open challenge and requires robust mechanisms to enforce correct behavior during runtime. This paper illustrates a runtime safety enforcement framework for the output sanitization of an autonomous driving agent on a highway. The enforcement mechanism is based on a (formally validated and verified) Asmeta model representing the enforcement rules and used at run-time to eventually steer the driving agent to behave safely and avoid collisions. We demonstrate both efficacy and efficiency of the proposed enforcement approach by conducting an experimental evaluation. We connected our safety enforcer with the highway simulation environment and co-executed it with the pre-trained (unsafe) AI agents as provided by the ABZ 2025 case study. We consider the single-lane case with the safety requirement and one scenario of the multi-lane case about preferring the right-most lane.