The escalation of vehicular fatalities globally has necessitated the transition from passive safety measures to proactive, predictive Advanced Driver Assistance Systems (ADAS). This research proposes a novel Artificial Intelligence-Integrated Mechanical Safety System (AI-IMSS) designed to predict potential collisions and autonomously actuate mechanical safety protocols with sub-millisecond latency. The proposed architecture fuses a Long Short-Term Memory (LSTM) network for temporal trajectory prediction with an Extended Kalman Filter (EKF) for robust state estimation under Gaussian noise. By analyzing vehicle kinematics, driver behavior, and environmental variables, the system calculates a dynamic Risk Score (Rs). Upon exceeding a critical threshold, the system triggers a dual-stage actuation: pre-tensioning of safety restraints and Autonomous Emergency Braking (AEB). Experimental validation using the CARLA simulation environment demonstrates that AI-IMSS achieves a collision prediction accuracy of 98.4%. Statistical analysis via paired t-tests reveals a significant reduction in reaction time (p < 0.001) compared to standard driver responses, with a Cohen’s d effect size of 2.14. These results substantiate the system's efficacy in mitigating accident severity in high-speed, stochastic traffic scenarios.