View of Instruction Alignment and Risk Calibration in Large Language Models for Safe Human-AI Interaction | Journal of Computer Technology and Software