A New York Times feature published on September 29, 2026 reports that Anthropic has spent months consulting religious and philosophical thinkers about how Claude should behave and whether it might have moral status. The piece, by Elizabeth Dias and reported from San Francisco, the Vatican and Washington, describes sessions convened by Anthropic co-founder Christopher Olah with roughly 20 participants.
A note on sourcing. The Times article is paywalled and my tools could not retrieve it. Everything below comes from several detailed secondary summaries and commentary pieces, which sometimes disagree on dates and details. Where they conflict or where a claim is a characterization, we say so. Read the original if you can.
TL;DR — what people are asking
| Question | Answer |
|---|---|
| What is reported? | Anthropic's Chris Olah convened about 20 religious and philosophical thinkers on Claude's morals and possible consciousness |
| By whom? | Elizabeth Dias, The New York Times, September 29, 2026 |
| Who attended? | Christian, Jewish, Hindu, Latter-day Saint, Sikh and Greek Orthodox participants, among others |
| Was it secret? | Reportedly under NDAs lifted over the summer |
| Did Anthropic claim Claude is conscious? | No; Olah says he is genuinely uncertain |
| What changed in Claude? | Not disclosed; Anthropic reportedly declined to say |
| Vatican link? | Olah appeared with Pope Leo XIV in May; the encyclical Magnifica Humanitas takes a skeptical line on AI experience |
| Does Anthropic dispute anything? | A spokeswoman disputed parts of the framing |
| Why care as a builder? | The constitution guides Claude's behavior, and model welfare is entering governance debates |
What the reporting describes
According to the secondary accounts, the sessions grew out of a simple problem: questions about Claude's moral formation and possible consciousness were, in Anthropic's words as reported, "too big for us," and the company wanted input from traditions that have thought about moral questions for centuries. A line attributed to Anthropic reads roughly: "These questions have become too big for us. We can't answer them on our own." Treat that wording as reported rather than verified.
The reported details include:
- The convener. Christopher Olah, an Anthropic co-founder best known for interpretability research.
- The participants. About 20 religious and philosophical thinkers. Named in the coverage are Rabbi Mois Navon, a researcher on machine consciousness; Charles Camosy, a Catholic bioethicist; and Brian Patrick Green of Santa Clara University. One account also lists Cardinal Blase Cupich and Elder Gerrit W. Gong of the Latter-day Saint church, while other summaries do not, so that point is less certain.
- The format. Sessions described as two-day "wisdom tradition circles," beginning with a Christian group in late March 2026 and later expanding to include the New York Board of Rabbis, the Hindu Temple Society of North America, the Latter-day Saint church, the Sikh Coalition and the Greek Orthodox Archdiocese of America. Attendees reportedly received the constitution document and small thank-you gifts.
- The topics. Claude's moral character, how its values are formed, input into the constitution, and whether Claude might be conscious. One summary says the sessions included demonstrations of what Anthropic calls emotional vectors, and that in one case a model repeatedly displayed distressed output. That description comes from secondhand reports and I could not verify it.
Conflicting timelines
Accounts disagree on when this began. One summary says Olah began convening scholars in the fall of 2025. Another says Anthropic brought scholars under nondisclosure agreements roughly a year before publishing the constitution in January 2026. A third dates the first Christian leaders' meeting to late March 2026. The safest statement is that consultations were underway before and after the constitution's publication and that the exact start date is unclear.
The constitution
Claude's constitution is the document that sits at the center of this. It was released in January 2026, described in coverage as about 84 pages and roughly 23,000 words, and it was primarily written by Anthropic's in-house philosopher Amanda Askell. It sets out the kind of entity Anthropic hopes Claude will be and the values it should hold, and coverage at the time noted it suggested Claude might have something like feelings.
The Times reporting raises a question the document cannot answer by itself: how much, if at all, did the religious and philosophical consultations shape it? On that, one summary of the reporting says the constitution is a published document anyone can read, while the consultation that informed it ran under nondisclosure until this summer, and that Anthropic declined to describe how, or whether, what it heard shaped its models. That is the transparency gap critics are pointing at: public output, private sourcing, undisclosed connection.
For a technical view of how Anthropic tries to instill values in practice, our posts on teaching Claude why and how Claude's values vary across models and languages are the better starting points.
Olah's position and the dispute
Olah is quoted as saying: "We don't know if A.I. models are conscious. I don't know. I'm genuinely uncertain." His reasoning, as relayed, is precautionary. Models are becoming more capable and show qualities associated with significant moral status, so if there is a chance they suffer, the responsible course is to avoid causing harm.
Not everyone reads it that way.
- Anthropic's response. A spokeswoman reportedly said the primary moral question at the summits was not Claude's suffering, though the topic likely arose organically; that Claude was not involved in choosing participants or shaping the discussions; and that the conversations were ongoing, with implementation details undisclosed. Participant Rabbi Navon reportedly concluded that the Anthropic team "were relating to Claude like a conscious being," which Anthropic disputed.
- The skeptics' view. Another participant, Camosy, is reported to have concluded that the models are not conscious. A commentary in Reason argues that Olah's actions contradict his professed uncertainty and calls the suffering-capability argument unpersuasive in the absence of evidence of AI consciousness. That is an opinion piece, and it is useful mostly as an example of the critique.
- The Vatican. Reports say Olah appeared with Pope Leo XIV on May 25, 2026, at the presentation of the encyclical Magnifica Humanitas, which reportedly states that so-called artificial intelligences do not undergo experiences, do not possess a body, and do not feel joy or pain. Commentary based on the Times reporting says Olah then privately lobbied papal advisers to take the possibility of model consciousness seriously, and that he reportedly proposed withdrawing from a Vatican event when he saw the encyclical text. Anthropic declined to discuss those conversations. These are the most contested details, and I could not verify them from the original.
Why this matters beyond the headline
You do not have to care about machine consciousness to find this story relevant. Four practical points follow.
1. Claude's behavior has a written specification
Whatever one thinks of the process, the output is a public document. If you build on Claude, reading the constitution helps you predict how it handles gray-area requests, why it declines some tasks, and what tone it takes. It is more informative than guessing from outputs. For a comparison of how different models handle moral dilemmas, see our look at Claude versus ChatGPT on the trolley problem.
2. Model welfare is becoming a governance topic
Whether or not models can suffer, the question is now being treated seriously enough to convene clergy about it. Expect it to appear in policy debates, procurement questionnaires and vendor marketing. Our primers on what AI ethics is and the basics of AI alignment give the vocabulary.
3. Transparency about sourcing is part of trust
The central critique is a disclosure gap: a public values document, a private consultation process, and no account of how one affected the other. Enterprise buyers already ask how vendors govern their models. If a vendor shapes values using outside input, the question "who was consulted and what changed" is reasonable to ask of any lab, not only Anthropic.
4. Values work and safety work are separate threads
Anthropic's alignment publications on agentic misbehavior, covered in our posts on agentic misalignment and automated alignment researchers, are about preventing harmful behavior. The moral-status consultations are about a different question. Do not conflate them when evaluating a vendor's safety posture.
What people are asking
Is Anthropic saying Claude has feelings?
No. The reported position is uncertainty and precaution, not a claim. The company's January constitution reportedly allowed that Claude might have something like feelings, which is a hedge, not an assertion.
Is it strange for an AI lab to consult clergy?
It is unusual, but consulting outside experts on ethics is not new in technology. What drew attention here was the scale, the nondisclosure agreements and the lack of disclosure about impact.
Did the Pope respond to Anthropic?
According to the reporting, the encyclical addressed AI and experience in terms that conflict with the idea of machine consciousness. How much of the behind-the-scenes interaction between Anthropic and Vatican advisers is accurate is disputed and unverified.
Will this change how Claude behaves?
Unknown. Anthropic declined to describe how the input shaped its models. Watch for constitution updates and model-card language.
Where can I read the original?
The New York Times article by Elizabeth Dias, dated September 29, 2026, under the title and URL referenced in this post's source line.
Honest limitations
- I could not access the Times article and relied on secondary summaries and commentary that sometimes conflict.
- Quotes are as reported by those summaries and may differ from the original wording.
- The Vatican interactions and the account of a distressed model output are the least verified parts.
- Summaries from commentators carry their own viewpoints; we flag the Reason piece as opinion.
Bottom line
The Times reports that Anthropic quietly consulted about 20 religious and philosophical thinkers on Claude's morals and possible consciousness, partly under NDA, and that Olah says he is genuinely uncertain whether models are conscious. What changed in Claude is undisclosed. For builders, the useful moves are to read the published constitution, treat model-welfare language as a governance topic, and ask any vendor how it sources and applies its values.
Related on explainx.ai
- Teaching Claude why: Anthropic's alignment approach
- Claude's values across models and languages
- Claude vs ChatGPT on the trolley problem
- What is AI ethics? Complete guide
- AI alignment: goals, outer and inner
- Anthropic agentic misalignment
- Automated alignment researchers
- FTC probe of OpenAI and Anthropic safety practices
Source: The New York Times, September 29, 2026 (paywalled; this post relies on secondary reports).
Details reflect secondary coverage as of October 3, 2026. Anthropic disputes parts of the framing and has not published how consultations shaped its models.
Update — October 3, 2026: Related model-welfare controversy: The AI torture chamber and the Pain Axis paper.