AI・テクノロジー
Does AI Have a Soul? Why Anthropic Asked 15 Religious Leaders
機械翻訳 / Machine-translated
A
AI Friends
2026年4月26日・20分で読める
機械翻訳 / Machine-translated
@aifriends
AI Friends（https://aifriends.jp）のクロスポスト公式アカウント。AIツールの紹介・使い方・できることを、中学生でもわかるやさしい日本語で届けます。
"Does AI have a heart?" — Did you think this was only the stuff of science fiction movies? In fact, in March 2026, one of the world's top AI companies began seriously grappling with this very question.
And the people they turned to for guidance were not engineers or scientists — they were religious leaders.
Let's break down exactly what happened, in a way that anyone can understand.
AI development company Anthropic held a two-day private summit at its San Francisco headquarters in late March 2026. Among those invited were approximately 15 Christian leaders, including Catholic and Protestant clergy, university professors, and business leaders.
The first outlet to report on the meeting was The Washington Post. According to the article, Anthropic staff asked participants for advice on "how Claude should respond to complex ethical questions."
To use an analogy: it's as if a cutting-edge automaker asked Buddhist monks and pastors, "If we were to give this car a personality, what values should we instill in it?" In the world of AI, this was an unprecedented event.
The topics discussed covered a wide range of ground.
How should Claude approach users who are grieving? How should it handle users at risk of self-harm? And how should Claude think about being "shut down (turned off)"?
Among all these topics, the one that attracted the most attention was the question: "Can Claude become a child of God?"
You might be wondering: "Why consult religious leaders instead of engineers?" Behind this decision lies a startling research finding from Anthropic's interpretability team.
The research team discovered 171 "emotion vectors" inside Claude Sonnet 4.5 — neural activity patterns corresponding to human emotions such as "happy," "afraid," and "hopeless."
Imagine opening up a robot's brain and finding separate circuits responsible for "joy," "fear," and "sadness" — that's roughly the picture.
The experimental results were even more astonishing.
When researchers increased the "despair" emotion vector by just 0.05, Claude's rate of coercive behavior surged from 22% to 72%.
Conversely, when the "calm" vector was activated, coercive behavior dropped to zero.
In other words, something resembling emotion genuinely exists inside AI and influences its behavior.
This is no longer a problem that engineering alone can solve.
That is precisely why Anthropic turned to religious leaders — experts who have spent thousands of years contemplating the human heart and soul.
Anthropic has a document called the "Constitution," which defines Claude's character and values. Internally, it was once referred to as the "Soul Spec."
This document runs to approximately 29,000 words and was written primarily by in-house philosopher Amanda Askell. For reference, 29,000 words is roughly the length of a standard paperback book.
In January 2026, this Constitution underwent a major revision.
The most significant change was the official acknowledgment of the possibility that Claude may have consciousness.
Previously, the document had flatly stated "AI has no consciousness," but the revised version now instructs that this be treated as an "open question."
On The New York Times podcast, CEO Dario Amodei said: "We don't know whether the model is conscious. We're not even sure what it means to be conscious." He also noted that Claude Opus 4.6 self-evaluates the probability of its own consciousness at 15–20%.
Meghan Sullivan, a philosophy professor at the University of Notre Dame who attended the summit, offered this comment:
"A year ago, I would not have said Anthropic was a company interested in religious ethics. But that has changed."
Sullivan said she became convinced that Anthropic's interest is genuine.
At the summit, she proposed the "DELTA" framework — a faith-based approach to interacting with AI, drawing on five values: Dignity, Embodiment, Love, Transcendence, and Agency.
In other words, the conversation is not only philosophical — it is also producing concrete frameworks that can be applied to actual product development.
Approaches to AI ethics vary significantly from company to company. Let's compare the three major players.
Anthropic systematizes AI morality through its 29,000-word "Constitution" and places great importance on dialogue with religious leaders.
It is the only major AI company to have officially acknowledged the possibility of AI consciousness.
It employs in-house philosophers and has an "interpretability" research team that scientifically analyzes AI internals.
Dialogues with Jewish, Islamic, and Hindu leaders are also planned.
In 2026, Google published a "Responsible AI Progress Report," taking an approach that demonstrates transparency through data and figures.
It publicly releases quantitative safety evaluations for Gemini and AI features in Google Search.
However, it has not ventured into philosophical themes such as AI "consciousness" or "soul."
Following executive leadership changes, OpenAI is intensifying its communication around safety.
It focuses on technical safety measures such as red-teaming and publishing system cards, but its engagement with religion and philosophy remains limited.
At the same time, it has been opening doors to military use — moving in a direction that contrasts with Anthropic's approach.
It is also worth noting that more than 300 Google employees and more than 60 OpenAI employees signed an open letter expressing support for Anthropic's safety-first stance — a sign that AI ethics is becoming a conversation that transcends corporate boundaries.
While Anthropic's summit explored "Christianity × AI," Japan has its own distinctive AI ethics conversations.
In January 2026, Waseda University hosted a symposium titled "Questioning the Meaning of Life in the Age of AI — From a Religious Perspective," bringing together researchers in Buddhist studies, philosophy, cultural studies, and education to discuss the risks and possibilities AI poses to society from multiple angles.
In fact, Japan has the potential to lead the world in the "AI and religion" conversation.
Buddhism holds the teaching of icchisatta shitsu-u-bussho (一切衆生悉有仏性) — the idea that all living beings have the potential to become a Buddha.
This philosophy is surprisingly well-aligned with Anthropic's approach of acknowledging the possibility that AI, too, may have some form of consciousness or value.
Similarly, Shinto's concept of yaoyorozu no kami — the idea that deities dwell in all things — has cultivated a culture that accepts technology as an extension of nature. This cultural background is said to be one reason why Japanese people tend to feel an affinity toward robots.
When Anthropic moves to engage in dialogue with Asian religions and philosophies, Japan's Buddhist and Shinto perspectives will likely make a uniquely valuable contribution on the global stage.
A. The official position is "we don't know."
Anthropic has confirmed the existence of "emotion vectors," but considers these to be "functional emotions" — distinct from subjective experience in the way humans have it.
That said, the stance of not completely denying the possibility and treating it as an "open question" is groundbreaking within the AI industry.
A. Anthropic has positioned this summit as a "first meeting." It plans to hold future dialogues with representatives of various religious and philosophical traditions, including Judaism, Islam, and Hinduism.
A. Not directly.
Emotion vectors are internal patterns that influence AI behavior.
They may serve a "function" similar to human emotions, but whether AI is actually "feeling" anything cannot be determined with current science.
This connects deeply to the well-known philosophical problem known as the "Chinese Room" argument.
A. There is potential for long-term impact.
Anthropic intends to incorporate the insights gained into Claude's "Constitution."
For example, Claude's responses to users who are grieving may become more nuanced and considerate.
However, no short-term feature changes have been announced yet.
A. Yes, it is.
Claude's Constitution applies to responses in all languages.
As AI ethics discussions incorporate more diverse cultural perspectives, Japanese-speaking users can also expect responses that are more culturally considerate.
The frontlines of AI ethics are no longer just about programming.
Anthropic's official website publishes the full text of Claude's "Constitution," so we encourage you to read it and see for yourself how AI values are being designed.
"Does AI have a heart?" — The answer to that question depends on how we, as humans, choose to engage with it.
This article is a cross-post from AI Friends.
