Risk Awareness Injection: Calibrating Vision-Language Models for Safety without Compromising Utility Paper • 2602.03402 • Published Feb 3 • 1