“Last week, Anthropic proposed a system where a government body could work with developers to pause AI models if they become too dangerous.”