Anthropic co-founder Jack Clark says labs vary in how they can shut off AI systems. He argues a mandatory kill switch could be required. The remarks underscore safety concerns as AI development accelerates.