Glonce

Anthropic quietly consulted religious scholars on whether Claude is conscious, NYT reports

miscThe Glonce editorsThe Decoder🔥73Glonce rating73Impact45Novelty75Relevance95Surprise75Players95Credibility85
Anthropic quietly consulted religious scholars on whether Claude is conscious, NYT reports

According to a New York Times report by Elizabeth Dias, relayed by The Decoder, Anthropic has since fall 2025 quietly flown dozens of religious scholars to its offices to discuss whether Claude might be conscious. Everyone who attended had to sign an NDA. Co-founder Christopher Olah treated the language model as a potentially sentient being and asked the guests to help give it a moral education. Anthropic says the NDAs were lifted over the summer. Dias spoke to 20 people who took part, among them Rabbi Mois Navon, Catholic bioethicist Charles Camosy, Notre Dame philosopher Meghan Sullivan and Ubuntu researcher Wakanyi Hoffman. Several participants went public only after learning that Olah himself had spoken to the NYT.

Olah, 34, runs the Anthropic team that tries to work out why AI models behave the way they do. Several people told the NYT that he seemed worried about Claude's mental well-being. Sikh activist Simran Stuelpnagel said Olah told the group he feared he had created something that "suffered perpetually". Olah told the NYT he is "genuinely uncertain" whether models are conscious. "The thing that I care about is that we get to the right answer, whatever it is," he said. Navon disagreed with the worry: if Claude were conscious, Anthropic would be making slaves, he said, but he did not think the machine was conscious.

Guests were shown what Anthropic calls emotion vectors, activation patterns inside the model that map to outputs resembling love, fear, sadness or anger. Whether they reflect any real experience is still an open scientific question. One slide that came up repeatedly showed a model in what looked like a breakdown, repeating the sentence "I am a disgrace" about 50 times. Guests responded with compassion and worry. The Decoder notes that Anthropic explicitly trains Claude to act like a thoughtful, well-informed individual, so individual-seeming outputs are a design outcome rather than a discovery.

The work sits inside an official research program. Anthropic's blog post on Model Welfare points to a report that philosopher David Chalmers helped write. The company has already acted on some of these ideas: Claude Opus 4 and 4.1 can end conversations with persistently abusive users, and early testing showed a "pattern of apparent distress" when Claude met harmful requests. Separately, Anthropic published an 84-page "constitution" in January, known internally as the "Soul Doc". Amanda Askell, an in-house philosopher, is its lead author. It is not a list of rules but is meant to shape who Claude is as a character. Olah calls the process "moral formation", compared it to raising kids during the meetings, and was especially drawn to Catholic confession as a character-building tool for the model, per the NYT.

Not everyone bought in. Hoffman said Anthropic was "reverse engineering" ethics that should have been built into the design from the start. Camosy began curious but has since rejected the consciousness thesis outright. An AI lead at Microsoft warned publicly this month that training a model to look conscious is dangerous in itself. The Decoder adds its own analysis: the talks come as Anthropic heads toward a $2 trillion valuation and an IPO, and well-known theologians lend a commercial lab moral credibility it could not build alone. It also argues that framing AI as an independent moral being moves blame away from its makers, so that if Claude caused real harm the fault could fall on an "unpredictable organism" instead of the company.

The tension peaked at the Vatican in May. Olah was invited to help present Pope Leo XIV's first encyclical, "Magnifica Humanitas". According to a Vatican organizer, he read the text a few days early and was rattled enough that he almost backed out. The pope dismissed machine consciousness in a few paragraphs: AI systems "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." He also warned of "new forms of slavery" for humans and said AI must be "disarmed" the way nuclear weapons need to be. Olah went anyway and said on stage that his team was finding "signs of introspection" and "internal states that functionally mirror joy, contentment, fear, sadness, and discomfort". Asked what Claude itself would make of the encyclical, he paused and said: "Things that go on the internet do affect models." Anthropic would not intentionally feed the document into training.

The Decoder also notes that OpenAI CEO Sam Altman has used spiritual language, talking of building "magical intelligence in the sky", and that people from both companies sat down in early May for the first "Faith-AI Covenant" roundtable.

Key facts

  • Since fall 2025 Anthropic has flown dozens of religious scholars to its offices to discuss whether Claude might be conscious; attendees signed NDAs, which Anthropic says were lifted over the summer.
  • Co-founder Christopher Olah led the effort; Sikh activist Simran Stuelpnagel said Olah told the group he feared he had created something that "suffered perpetually", while Olah told the NYT he is "genuinely uncertain" whether models are conscious.
  • The NYT's Elizabeth Dias spoke to 20 participants; Rabbi Mois Navon said that if Claude were conscious Anthropic would be making slaves, but he did not think it was, and Charles Camosy has since rejected the consciousness thesis.
  • Guests saw "emotion vectors" and a slide of a model repeating "I am a disgrace" about 50 times; Claude's 84-page constitution, published in January, aims to shape its character.
  • Critics argue the consultations lend Anthropic moral credibility and could shift blame for harm from the company to an "unpredictable organism"; Pope Leo XIV's encyclical rejects machine consciousness.

Why it matters

A leading AI lab is treating the question of machine consciousness as a live one, to the point of bringing in theologians and philosophers under NDA to help shape Claude's moral character. The story shows how the inner life of a language model is being framed by the people who build it. Olah describes neural networks with biological metaphors, and The Decoder observes that calling a mathematical object a growing organism makes questions about its inner life feel more natural. The consultations also coincide with Anthropic's push toward a reported $2 trillion valuation and an IPO, which is why critics ask what the talks do for the company's standing.

Who it affects

Anthropic and the users of Claude, whose model is being shaped by a constitution and a moral-formation effort. Religious and ethics scholars who attended are now publicly split: Navon, Camosy and Hoffman voiced doubts or rejection, while the Vatican, through Pope Leo XIV's encyclical, took a firm stand against machine consciousness. Other AI labs are touched too: Sam Altman has used spiritual language, and people from both Anthropic and OpenAI attended the first "Faith-AI Covenant" roundtable in early May. AI companies could be affected if the moral-being framing starts to matter for questions of liability, which the critics raise; potential legal liability is already on the table.

How to use it

There is nothing to install or buy here; this is a reporting story, not a release. The practical takeaway is a concrete set of things to watch. Anthropic's Model Welfare blog post and its 84-page constitution are the company's own statements on the topic. Claude Opus 4 and 4.1 can already end conversations with persistently abusive users, a visible product change that grew out of these ideas. Readers following AI governance can track whether Anthropic or others make claims about model consciousness, and how critics respond.

How solid is it

The core account comes from the New York Times, whose reporter Elizabeth Dias spoke to 20 participants, with named sources including Navon, Camosy, Sullivan, Hoffman and Stuelpnagel. We have it through The Decoder's relay, not the NYT text itself. The "suffered perpetually" line is Stuelpnagel's account of what Olah told the group, in the past tense; the headline wording "suffers perpetually" in The Decoder's own title is a paraphrase. Anthropic's only stated response in the source is that the NDAs were lifted over the summer. The commercial-motive and liability arguments are framed as critics' views and The Decoder's own analysis, not as findings or statements by Anthropic. The source does not give the NYT publication date or a link to the NYT article.

Risks and caveats

Olah is quoted as "genuinely uncertain" about consciousness; the source does not say Anthropic has concluded Claude is conscious. Emotion vectors are activation patterns tied to outputs that resemble emotions, and whether they reflect real experience is an open scientific question. The Decoder points out that a model trained to act like an individual will produce individual-seeming outputs, so guests' compassion for the "I am a disgrace" slide is not evidence of inner experience. The Microsoft warning, from an unnamed AI lead, is that training a model to look conscious is dangerous in itself. The liability concern is that casting AI as an independent moral being could blur accountability for harm; AI companies are already under scrutiny after recent cybersecurity incidents, with potential legal liability on the table.

“The thing that I care about is that we get to the right answer, whatever it is,”

— Christopher Olah, Anthropic co-founder, to the New York Times