Driven by AI, the concept of MTD has gained attention as an effective means to strengthen the protection of complex and dynamically evolving cyber-physical infrastructures. One notable application scenario is the SD-IoV, a critical enabler of the broader IIoT. Traditional security strategies often operate reactively—only responding after intrusions—which weakens their capability against adaptive and sophisticated cyber threats. To overcome these limitations, this work presents a novel AI-enhanced MTD architecture that adaptively reconfigures network states to obscure system exposure and reduce vulnerability in SD-IoV settings. Compared with existing MTD methods that lack flexibility and exhibit low autonomy, our design leverages DRL to autonomously explore and select optimal reconfiguration policies, moving beyond fixed defense-response schemas. In our design, the operation logic of RSUs is abstracted as a MDP, which empowers them to make context-aware decisions in fluctuating environments. Moreover, we incorporate a vehicle-level trust assessment component, which helps detect and isolate suspicious participants—such as compromised or surveillance vehicles—post adaptation. Simulation-based validation reveals that the proposed AI-oriented MTD mechanism significantly outperforms standard approaches in countering DDoS attacks, showcasing its practical advantage in proactive defense.