Forsberg, L. Instruction Alignment and Risk Calibration in Large Language Models for Safe Human-AI Interaction. JCTS 2025, 4.