这篇文章聚焦梵蒂冈与 Anthropic 之间的交集：教宗 Leo XIV 在 2026 年的 AI 通谕《Magnifica Humanitas》中呼吁对 AI「解除武装」，而 Anthropic 共同创办人 Chris Olah 竟在教廷活动上发言。文中指出，Olah 既是无神论者、又曾是 Thiel fellow，却承认包括 Anthropic 在内的前沿 AI 实验室都受奖励与约束牵引，可能与「做正确的事」冲突；这也呼应教宗对 AI 需要外部压力与内部节制的判断。
文章强调，教宗并非指望通谕立刻阻止 AGI 竞逐、裁员或 AI 武器化，而是要透过对话迫使业界放慢鲁莽野心，甚至产生某种羞愧感。梵蒂冈关注 AI 已有多年：自 2016 年起举办 Minerva Dialogues，2015 年又邀请科技人物参与相关讨论；到 2025 年，圣荷西的天主教伦理学者与神职人员主动接触矽谷 AI 圈，最终把 Olah 视为重要内部人。Anthropic 的 Claude 宪章更新也反映了这些对话，Olah 曾把草案交给 Green 与 McGuire，后者回传长达 28 页的意见。
然而，Olah 与 Leo 的世界观并不完全一致。Olah 认为模型比科幻作品想像得更细腻、古怪且美丽，并暗示它们可能朝近似人类的地位逼近；相对地，Leo 在通谕第 99 段明确警告，不应把这种「智能」等同于人类智慧，并批评 transhumanism 的「人机混合」想像。文末指出，AI 是否是「工具」还是某种新型存在仍未定论，但伦理风险已迫在眉睫；借由 Anthropic 这位盟友，教宗已为 AI 工业内部展开更尖锐的道德对话提供了基础。
This article examines the overlap between the Vatican and Anthropic: in Pope Leo XIV’s 2026 AI encyclical, Magnifica Humanitas, he calls for “disarming” AI, while Anthropic cofounder Chris Olah unexpectedly spoke at a ceremony tied to it. Olah, an atheist and former Thiel fellow, acknowledged that frontier AI labs, including Anthropic, work within incentives and constraints that can conflict with doing the right thing, reinforcing Leo’s view that AI needs both external pressure and internal restraint.
The piece argues that Leo does not expect the encyclical to immediately stop AGI races, layoffs justified by AI efficiency, or AI weapons programs. Instead, its aim is to foster dialogue that could slow reckless ambition and perhaps create shame among builders who fear the outcome may be disastrous. The Vatican has been building this conversation for years: it launched the Minerva Dialogues in 2016, invited tech leaders, and by 2025 Catholic ethicists and clergy in San Jose had begun engaging the local AI industry, eventually drawing Olah into the role of insider witness. Anthropic’s Claude constitution update also shows the influence of those discussions, including a 28-page response from pastor Brendan McGuire and acknowledgments for both Brian Patrick Green and McGuire.
Still, Olah and Leo diverge on a central philosophical issue. Olah describes AI models as subtle, odd, and beautiful, “made from us, from our words,” implying a possible path toward humanlike status; Leo, in paragraph 99 of the encyclical, explicitly rejects equating machine intelligence with human intelligence and criticizes transhumanism’s “human machine hybrid” ideal. The article concludes that this debate will not be settled soon, but the moral questions around AI must be addressed now, and Leo’s alliance with an Anthropic figure gives the church a foothold for harder conversations inside the AI world.