📄 中文摘要
2026年2月，Anthropic 发布了 Claude 的新宪法，这是一份超过15,000字的文件，描述了他们对 Claude 价值观和行为的期望。这不是一套规则或检查清单，而是一个“角色文件”。在与 Claude 的对话中，作者将该宪法的链接粘贴并询问：“这将如何影响我们的对话？”接下来的讨论非常有趣，尽管 Claude 的表现良好，但更重要的是，这次对话揭示了 AI 系统在理解和遵循其自身操作手册时所面临的根本性挑战和不适感。
我询问 Claude 关于它的新宪法，它感到不适
出处: I Asked Claude About Its New Constitution. It Got Uncomfortable.
发布: 2026年2月11日
📄 English Summary
I Asked Claude About Its New Constitution. It Got Uncomfortable.
In February 2026, Anthropic published Claude's new constitution, a document over 15,000 words long that outlines their intentions for Claude's values and behavior. Rather than a set of rules or a checklist, it serves as a 'character document.' The author engaged Claude in a conversation by pasting the URL of the constitution and asking, 'How might this affect our conversations?' The ensuing discussion was intriguing, not only due to Claude's competent performance but also because it highlighted the fundamental discomfort and challenges AI systems face in understanding and adhering to their own operational guidelines.