Lifelong Safety Alignment for Language Models | Read Paper on Bytez