JailBound: Jailbreaking Internal Safety Boundaries of Vision-Language Models | Read Paper on Bytez