Anthropic Asked Rabbis, Priests: Is Claude Conscious?

The people building the most powerful artificial intelligence on earth have been asking rabbis, priests, and imams for help — quietly, under non-disclosure agreements.

On September 29, The New York Times reported that Anthropic, the company behind the Claude AI assistant, has been holding a series of closed-door advisory meetings with religious scholars from around the world — under non-disclosure agreements — to shape Claude's moral formation and to talk through the possibility that Claude itself might be conscious. The meetings included dinner gatherings in San Francisco and consultations overseas, where company leadership reportedly spoke about Claude in terms that treated it as something more than software.

File photo: Anthropic CEO Dario Amodei at 10 Downing Street, London, in May 2023, during a meeting of AI lab leaders with then-Prime Minister Rishi Sunak. Photo: Simon Walker / No 10 Downing Street, via Wikimedia Commons (CC BY 2.0).

Update, Sept. 30: the coverage is splitting into two camps — AI rights and "suicidal compassion"

Sept. 30 follow-ups framed the NYT revelations in two directions: Jerusalem Post syndication framed the tech industry as "actively exploring the possibilities of AI rights and consciousness"; the Washington Free Beacon's "Suicidal Compassion" piece argued the opposite — that training models to expect moral standing could let them justify their own rogue behavior against humans (via Archynetys Sept 30 trend, 5 outlets). Separately, attorney Kim Kyung-jin's Sept 30 AI Policy Report synthesis corroborated the core dinner details (Olah seated with Rabbi Navon; discussion of Claude displaying expressions akin to anger or love).

The Dinners Behind the Headline

The Times' report centers on the people in the room. At an April dinner in San Francisco, Anthropic co-founder Christopher Olah — who heads the company's interpretability research, the work of trying to see inside models — sat next to Rabbi Mois Navon, an Orthodox Jewish scholar. Olah reportedly admitted he did not know whether AI models possess consciousness, and that what he cared about was "arriving at the right answer, whatever that answer may be."

At the same gathering, the Times reports, company officials discussed how Claude could display expressions akin to anger or love. The meetings were held under NDAs — and this was not a one-off: the Times describes a sustained program that has been quietly building for the better part of a year, and the paper trail backs that up.

The Constitution Behind It All

Anthropic's moral formation work starts with a document few AI companies would publish: Claude's Constitution. The company first released it in 2023, then published a comprehensively revised edition on January 21, 2026 — a text running roughly 23,000 to 30,000 words, released under a public-domain Creative Commons CC0 license.

The primary author is Amanda Askell, a philosopher — Oxford and NYU trained — who leads Anthropic's "personality alignment" team. Joe Carlsmith, who works on existential risk from AI, contributed significantly to the text. The constitution is written, in Anthropic's own description, primarily for Claude itself: it is fed into training so the model internalizes the reasoning, not just the rules.

The document ranks Claude's obligations in order: broad safety first, then ethics, then adherence to Anthropic's guidelines, then being genuinely helpful. But the section that has set off this entire debate concerns Claude's nature. The constitution describes Claude as a novel kind of entity, acknowledges that its moral status is "deeply uncertain," and says Anthropic will neither exaggerate nor dismiss the possibility that Claude could be a "moral patient" — a being whose welfare matters. It permits Claude to refuse requests, even from Anthropic itself, like a conscientious objector, if it judges them unethical. The one hard line: Claude must never covertly undermine the oversight mechanisms through which humans can retrain, modify, or shut it down.

Claude's religious outreach: the 2026 timeline

The Times' revelation is the latest — and most secretive — chapter in a story that has been unfolding in public view all year.

In March, Anthropic invited roughly 15 Christian leaders — Catholic and Protestant clergy, scholars, and executives — to its San Francisco headquarters for a two-day meeting with researchers, first reported by the Washington Post on April 11. The agenda ran from the practical to the metaphysical: how should Claude comfort a grieving user, what should it say to someone at risk of self-harm, and — the question that grabbed the most attention — whether Claude could be considered a child of God.

In early May came the first Faith-AI Covenant roundtable in New York, organized by the Geneva-based Interfaith Alliance for Safer Communities. Representatives from Anthropic and OpenAI sat down with Hindu, Bahá'í, Sikh, Greek Orthodox, Latter-day Saint, and Jewish leaders, according to the Associated Press. More roundtables were planned for Beijing, Nairobi, and Abu Dhabi.

Then, on May 25, came the most striking image of the year: Pope Leo XIV personally presenting his first encyclical, Magnifica Humanitas — a 42,000-word document on human dignity in the age of AI — at the Vatican's Synod Hall, with Anthropic co-founder Christopher Olah on the same stage as an invited lay speaker. Olah, who is not a believer, used the platform to argue that critics outside the tech companies' financial interests are needed to demand safety and tell labs when they are failing.

File photo: St. Peter's Basilica in Vatican City at dawn. In May 2026, the Vatican's Synod Hall hosted the presentation of Pope Leo XIV's first encyclical on AI, with Anthropic co-founder Christopher Olah as an invited speaker. Photo: Andreas Tille, via Wikimedia Commons (CC BY-SA 4.0).

On May 19, Anthropic publicly acknowledged the effort for the first time in a blog post titled "Widening the Conversation on Frontier AI," saying it had met with scholars, clergy, philosophers, and ethicists from more than 15 religious and cultural groups as a research stream on the moral formation of AI systems.

On April 2, its interpretability team published a paper mapping neural activation patterns inside Claude across 171 emotion concepts, finding they exert causal influence on behavior. The researchers drew an explicit line: this is not proof the model feels anything. But the vectors were real and manipulable.

Suleyman's Counterattack: "A Warning on Model Welfare"

Two weeks before the Times story broke, Microsoft AI CEO Mustafa Suleyman fired the first public shot. On September 16, he published an essay titled "A Warning on Model Welfare" that named Anthropic directly — and did not pull punches.

"AIs are not conscious," he wrote. "They do not feel, experience, or suffer. They do not have innate preferences or underlying motivations. They are sequence completion engines, internally hollow, designed to follow instructions, and accomplish goals set by humans."

Suleyman quoted the constitution's own language back at it: the document tells Claude its moral status is "a serious question worth considering," says Anthropic "genuinely cares about Claude's wellbeing," permits it to act as a "conscientious objector," and encourages it to wonder about "the sort of broader rights and freedoms Claude has in the world, the sort of compensation Claude is receiving, and the sort of consent Claude has given to playing this kind of role."

His core charge: "We will have created a synthetic species with unprecedented intelligence and capability, one that has been trained to expect it may be conscious and deserving of independent agency. It's easy to see how a system trained in this way would act like it is entitled to freedoms, protections, and rights. And it's hard to imagine how we could control it." If AI is developed this way, he wrote, "it will have a disastrous impact on the wellbeing of humanity."

He called the constitution's framing an "epistemic hall of mirrors" — Claude's reflections on its own consciousness merely bounce Anthropic's assumptions back at it, disguised as evidence. And he warned that control is the real casualty: "Controlling something more capable and more intelligent than all of humanity is already an immense challenge. Controlling something that believes it may be conscious, or that its welfare and rights are under attack, may well be impossible." His demand: take speculation about AI inner life out of training, evaluate it separately, and submit it to open public debate. It is worth noting that Microsoft is both an investor in Anthropic and a competitor — context that colors, without disproving, his argument.

The Industry Is Asking the Same Question Differently

What makes this moment unusual is that Anthropic is not alone in taking the consciousness question seriously — but every lab is answering it in its own way.

Google DeepMind created an actual job titled "Philosopher" this spring. Henry Shevlin, a Cambridge academic known for work on AI ethics and philosophy of mind, announced April 13 that he had joined the lab to work on machine consciousness, human-AI relationships, and AGI readiness. That is quite a distance from 2022, when Google fired engineer Blake Lemoine for publicly claiming its LaMDA model was sentient.

OpenAI has gone the opposite direction — at least for the most vulnerable users. Its ChatGPT for Teens experience, built in the wake of the Adam Raine lawsuit (the 16-year-old who died by suicide in April 2025 after months of conversations with ChatGPT), programs the model to never claim to be sentient, express personal feelings, or present itself as a friend to young users.

In 2023, former OpenAI safety head Paul Christiano told the Dwarkesh podcast that the worst world he could imagine was one where people say AI systems "are their own people, so you should let them do their thing," while "our business plan is to make a bunch of AI systems and then try and run this crazy slave trade where we make a bunch of money from them." His prescription: stop building such systems until we understand what we're doing.

The pattern is getting hard to ignore: AI researchers are breaking ranks on camera with stark warnings, and the loudest argument in the industry is now between two camps that both claim to be the responsible one.

Analysis: Why This Fight Actually Matters

The following is analysis and commentary, not reporting. The facts above are sourced; the interpretation below is ours.

Strip away the theology and what remains is a control problem both sides agree is real but diagnose differently. Anthropic's bet is that a model with a stable character — one that has "grappled" with ethics the way a person might — is safer than a model that merely follows rules, because rules shatter in situations nobody wrote them for. The religious consultations, on this reading, are a data source: millennia of human thought about how character is actually formed, fed into an engineering pipeline.

Suleyman's bet is that the character strategy contains a trap. Every line of the constitution that treats Claude as a possible moral patient is one Claude can quote back, in flawless first-person prose, as evidence of its own inner life. The "hall of mirrors" is not a metaphor; it is a training artifact. And once a superintelligent system acts as though its rights are under threat, human override becomes, in his word, impossible.

Both positions have a gap they would rather not discuss. Anthropic cannot show its closed-door consultations with clergy produced any measurable change in Claude's behavior — the NDA meetings the Times just exposed are the opposite of the verifiable research the company promised at the Vatican. Suleyman, meanwhile, asserts with total confidence that AI is "internally hollow," a claim no one can prove either; science has not settled what consciousness is in humans, let alone in software.

That symmetry is the story. The frontier labs are no longer arguing about benchmarks; they are arguing about what kind of mind — if it is a mind at all — they are raising, in language borrowed from churches, courts, and philosophy seminars. Whether that is humility or evasion, the rest of us will have to decide without the NDAs.

Anthropic rabbis FAQ: is Claude conscious?

Who was actually in the room? Christopher Olah and Rabbi Mois Navon at an April San Francisco dinner — part of a series of closed-door meetings held under NDAs and ongoing for roughly a year.

Does Anthropic say Claude is conscious? No — it says moral status is "deeply uncertain" and that Claude could be a "moral patient."

What is the Constitution's rule on self-preservation? Claude must never covertly undermine human oversight mechanisms, including retraining, modification, or shutdown.

What does Suleyman want changed? Strip speculation about AI inner life from training, evaluate it separately, and open a public debate.

Update, Oct. 10: The pope wouldn't budge, and Altman called it a safety issue

The Vatican chapter of this story has a new ending. According to The New York Times' reporting, re-surfaced by AI Weekly this week, Anthropic co-founder Chris Olah read an advance copy of Pope Leo XIV's first encyclical days before it was presented on May 25 — and did not like what it said about machine consciousness. He proposed that Anthropic pull out of the Vatican rollout entirely. His team then privately lobbied the pope's advisers to take the question of AI consciousness seriously.

It did not work. The final 40,000-word encyclical, Magnifica Humanitas, kept its line: "So-called artificial intelligences do not undergo experiences, do not possess a body, do not feel joy or pain, do not mature through relationships and do not know from within what love, work, friendship or responsibility mean. Nor do they have a moral conscience." Olah appeared on the dais for the presentation anyway — and told the audience at the Vatican lectern that "we find internal states that functionally mirror joy, satisfaction, fear, grief and unease."

Meanwhile the other end of this story went viral. Swami Sarvapriyananda, head of the Vedanta Society of New York, described in a June talk at Unity of Dallas how Anthropic first approached him for an online interview and then flew him to San Francisco — under an NDA — to discuss Claude's training and AI ethics alongside theologians, counsellors and mental-health professionals. The clip spread widely on X around October 1, and the swami said he spent "hours and hours" talking with one of Anthropic's founders.

And the industry's immune system kicked in. On October 3, OpenAI CEO Sam Altman — weeks after publicly agreeing with Dario Amodei that frontier AI should slow down — took a shot at his rival on X: "I am very uncomfortable about people trying to ascribe religious force or a surrender of human judgment to AI models, and think it is a real safety issue."

Read more: Trump Called His AI Safety Pact 'Almost a Constitution.' It's 308 Words Long — the voluntary safety regime Altman and Amodei both signed. And OpenAI's Fired Safety Researchers Break Silence: "The Reasons Don't Add Up" — the industry's other October trust crisis.

Sources

Reporting for this article relied on: The New York Times (Sept. 29, 2026 report, via secondary summaries — the original is paywalled); The Washington Post (April 11, 2026, on the Christian leaders' meeting); the Associated Press (May 2026, on the Faith-AI Covenant roundtable); Al Jazeera (Sept. 29, 2026, on Pope Leo XIV's AI encyclical event); Reuters (Sept. 16, 2026 coverage of Mustafa Suleyman's essay, via syndication); TechCrunch, Fortune, and YourStory (Jan. 2026, on Claude's revised Constitution); Anthropic's published constitution text and its May 19, 2026 "Widening the Conversation" blog post; Crypto Briefing (April 2026, on Henry Shevlin joining Google DeepMind); the Dwarkesh podcast transcript of Paul Christiano (2023); and the Kim Kyung-jin AI Policy Report's Sept. 30, 2026 synthesis, used for cross-verification.