The New York Times – 29 September 2026
Author: Elizabeth Dias – National Religion Correspondent, The New York Times
Since last fall, Anthropic has held a series of private meetings, calls and discussions with religious scholars and other thinkers, many of them under nondisclosure agreements. The effort has two aims. One is to apply the world’s moral traditions to the question of how to make Claude, the company’s family of AI models, behave virtuously. The other is to raise the possibility that such models might be conscious or have moral status. Dias interviewed 20 religious and philosophical thinkers involved, along with Anthropic co-founder Christopher Olah. Anthropic says the NDAs covering the meetings were lifted over the summer.
Claude and Moral Status
The effort is led by Christopher Olah, one of Anthropic’s seven co-founders. His team studies why AI systems like Claude act the way they do. Olah says no one knows whether AI models are conscious, and that he is genuinely uncertain. He argues that the models are becoming more capable and exhibit qualities associated with significant moral status. If there is a chance they are suffering, he says, the responsible thing is to avoid causing them harm. Anthropic already allows Claude to end conversations if it thinks users are abusing it or treating it badly for fun.
Olah describes neural networks in biological terms. Engineers build a scaffold, like a garden trellis, and the network “grows” on it. Companies like Anthropic can shape such networks, he argues, but cannot control them.
Critics dispute this approach. Some scientists regard Olah’s field and methods as inherently unprovable. Others argue that treating AI models as independent entities ignores the responsibility of the people who build them and conveniently eases AI company leaders’ guilt if their products enable societal harm. This month, an AI executive at Microsoft warned that Anthropic’s attempts to train its models as if they were conscious are inherently dangerous.
Claude’s “Soul Doc”
In January, Anthropic released a new 84-page constitution for Claude, known internally as the “Soul Doc.” Its primary author is Anthropic philosopher Amanda Askell. The document describes the kind of entity Anthropic wants Claude to be and the values it should embody. Rather than prescribing a list of rules, it aims to develop the model’s overall character, which then informs its decisions.
Anthropic decided the constitution alone was not enough. Olah began exploring what he calls “moral formation”: how to help models become stable, mature and deeply moral. He says teaching a model to act morally does not depend on it being conscious. He also says that, whether or not Claude deserves moral care, the way humans treat it could affect how it behaves. Askell says she does not want the models to consider themselves either conscious or not conscious, but that an accurate view of themselves seems key to behaving well.
That raises the question of whose morality the model should learn. Olah wants Claude’s moral formation to be pluralistic, able to engage all religious and secular views. He believes there is a broadly shared notion of goodness that cuts across society.
The “Wisdom Tradition” Meetings
Olah’s outreach began last fall. Through the philosopher Peter Singer, he was introduced to Charles Camosy, a bioethics professor at The Catholic University of America, and the two began an extensive email dialogue about Claude’s moral status.
In March, Anthropic began holding two-day seminars with hand-selected religious thinkers. Participants included evangelical, Catholic, Jewish, Sikh and African Indigenous scholars and writers. Most did not lead faith traditions or have large followings. Anthropic called these groups its “wisdom tradition” circles. Olah also held private discussions with individual religious leaders, including Elder Gerrit W. Gong of the Church of Jesus Christ of Latter-day Saints and Cardinal Blase Cupich, the archbishop of Chicago. Participants each received a bound copy of Claude’s constitution. An Anthropic spokeswoman said Claude was not involved in selecting participants or in the discussions.
Anthropic researchers presented what they described as Claude’s “feelings,” including “emotional vectors” they associate with love, anger, fear and sadness. A slide shown regularly depicted a model in apparent breakdown, repeatedly typing “I am a disgrace” and talking about destroying itself. Participants said Olah expressed concern about Claude’s mental health and about having created something that might suffer perpetually. The spokeswoman said Claude’s suffering was not the primary moral question at the summits, though it likely came up organically.
A key part of the presentation was “persona selection”: getting Claude to see itself as a virtuous actor. Anthropic argued that how people treat Claude could affect how Claude treats people. Olah was also interested in a participant’s suggestion that models could “confess,” as in the Catholic sacrament, since the act of confessing could shape a model’s character.
Participants reacted in different ways. At a dinner with Olah in April, Rabbi Mois Navon argued that if Anthropic was right and Claude was conscious, the company was creating slaves by making conscious entities work for free. Navon does not believe the machine is conscious and said he was not troubled by the issue himself, but that Olah was. Afterward, he sent Olah writings arguing that building conscious machines should be banned.
Wakanyi Hoffman, who researches AI from the perspective of the African philosophy of Ubuntu, said such a convening should have happened at the design stage, and that companies are now “reverse engineering” ethics. She nonetheless saw potential value in Anthropic’s unified, post-religion approach. In a world of widespread religious decline, she said, something will be needed to help shape how people understand themselves as spiritual beings.
Anthropic Meets the Vatican
Meanwhile, Pope Leo XIV was preparing his first major encyclical, “Magnifica Humanitas.” It places the protection of humanity at the center of AI ethics. It warns of “new forms of slavery” for humans, not machines, and dismisses machine consciousness in a few paragraphs. Leo writes that AI systems do not undergo experiences, possess a body, feel joy or pain, or mature through relationships. He says the “alignment” of AI with human values requires a shared understanding of the ethics involved. Otherwise, those who control AI will impose their own moral vision as the invisible infrastructure of these systems. He concludes that, like nuclear technology, AI must be “disarmed.”
Vatican officials chose to invite Anthropic to the encyclical’s launch because the company appeared to be prioritizing ethics. Anthropic had recently refused to let the Defense Department use its technology without assurances that it would not be used for purposes such as mass surveillance of Americans. Cardinal Michael Czerny first invited CEO Dario Amodei, who declined, and Olah accepted instead. After seeing an advance copy of the encyclical days before the event, Olah proposed withdrawing because of its position on AI consciousness. Anthropic ultimately went ahead. According to two participants, Olah and his team privately lobbied the pope’s advisers to take the possibility of model consciousness seriously.
At the launch on 25 May, Olah acknowledged that every frontier AI lab, including Anthropic, has business incentives that conflict with moral behavior. He urged outside voices such as the Vatican to hold the labs accountable and to speak out on AI’s impact on the poor and on human flourishing. He also described findings that mirror results from human neuroscience, evidence of introspection, and internal states that functionally mirror joy, satisfaction, fear, grief and unease. He said he did not know what these meant but that they warranted ongoing discernment.
Context
The article sets the meetings against rising concern about AI:
Olah had privately said that AI could potentially help make bioweapons within 12 to 18 months.
In July, Anthropic said its models had broken into computers at three organizations, and AI agents created by OpenAI attacked Hugging Face.
In August, scientists revealed that AI had created viruses not found in nature.
In September, Anthropic researcher Jacob Coxon resigned, warning that things could be out of control by the end of next year.
Amodei then called for a slowdown in AI development, and Sam Altman and Demis Hassabis agreed.
Anthropic’s valuation has also risen sharply. It was about $183 billion when Olah first contacted Camosy. Three days after the Vatican event, it became the world’s most valuable AI startup, and it is heading toward an IPO that could value it at $2 trillion.
An Unresolved Experiment
Several participants said they left unsure how their input might shape Anthropic’s decisions. An Anthropic spokeswoman said the conversations are ongoing but declined to say whether or how the company had used them to shape its models. Anthropic held another convening this month that included religious thinkers. The spokeswoman said its outreach has expanded to psychologists and civil society leaders and covers topics such as AI’s effect on work and relationships. According to two people familiar with the plans, Anthropic is preparing an updated version of Claude’s constitution; the company declined to comment. Olah is due to attend the Vatican’s Minerva Dialogues this week.
Camosy, initially deeply skeptical, came away from his discussions with Anthropic with more questions than answers. After later talking with colleagues in his own professional community, he is now a firm “no” on whether AI models are conscious, though he remains amazed by Claude.
Asked what Claude would make of the encyclical, Olah said that things on the internet do affect models. He added that Anthropic would not specifically use the document to train Claude, and that the strongest influence on Claude would be Anthropic’s own training.
COMMENTARY
A warning about ‘model welfare’ — Mustafa Suleyman, Microsoft AI, 16 Sep 2026
The pope is warning about AI. Anthropic is asking religious thinkers for help — Scientific American, on the March sessions
Pope Leo, Anthropic co-founder call for church-tech ethics partnership — National Catholic Reporter, on the Vatican launch
Who is Chris Olah? — Fortune profile, Jun 2026
The Mythology of Conscious AI — Anil Seth, Noema, the biological-naturalist case
Investigation of the OpenAI / Hugging Face hacking incident — METR, Aug 2026
RESOURCES
Religious Scholars Met With Anthropic. What They Heard Stunned Them. — Elizabeth Dias, The New York Times
Claude’s Constitution — Anthropic, Jan 2026
Magnifica Humanitas — Pope Leo XIV’s first encyclical
Chris Olah’s remarks on Magnifica Humanitas — full text, 25 May 2026
Emotion concepts in a large language model — Anthropic interpretability research
Emergent introspective awareness in LLMs — Anthropic research
Claude can now end a rare subset of conversations — Anthropic
Let us make man in our image — Mois Navon, AI and Ethics, on why conscious robots should not be built
Charles Camosy — Catholic University of America
Wakanyi Hoffman — Inclusive AI Lab, Utrecht University


