Aggressive driving often triggers anger and retaliatory behaviors, posing threats to traffic safety. This paper proposes an AI-driven apology mechanism based on an Augmented Reality Head-Up Display (AR-HUD), which delivers immediate apologies on behalf of offending drivers during traffic conflicts and repairs damaged social relations through prosocial lies. We conducted a 2 (scenario risk: high vs. low) × 5 (apology depth) mixed-design experiment (N = 40) to evaluate its effectiveness. Results show that AI apologies enhanced positive emotions and forgiveness intentions while reducing anger, with participants also perceiving psychological benefits. These effects were consistent across both high- and low-risk scenarios. Our findings offer a practical design pathway for human-AI emotional regulation in traffic contexts.