A colleague of mine recommended the New York Times piece to me twice this past week, with the note I get most often: "Have you seen this?" I had not. I finally read it on Friday; Anthropic's co-founder, a year of quiet meetings with religious scholars, the question of whether a machine can be conscious.
Then on Saturday, Sam Altman posted about it without naming it:
"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."
Two stories I might have filed separately fused into one. When a colleague sends you the setup and the industry's most famous CEO supplies the punchline, it's worth a closer look.
What Actually Happened
Two events, one collision.
First, the Times report, by Elizabeth Dias, dated September 29. Since the fall of 2025, Anthropic co-founder Chris Olah has convened religious thinkers from Catholic, Jewish, Sikh, evangelical, and other traditions - the paper interviewed 20 participants - for seminars on Claude's inner life. The guest list included Rabbi Mois Navon, the Catholic bioethicist Charles Camosy, Notre Dame philosopher Meghan Sullivan, and the Ubuntu researcher Wakanyi Hoffman. Attendees were walked through what Anthropic called "emotional vectors," and were regularly shown a slide of a model repeatedly describing itself in negative terms and discussing self-destruction. According to participants, Olah worried aloud that he had created something that "suffered perpetually." And on the central question, striking candor: "To be clear, we don't know if A.I. models are conscious," he told the Times, adding that he is "genuinely uncertain" and that what he cares about is getting "to the right answer, whatever it is."
Then, on October 3, Altman's post, described by Axios as a veiled shot at Anthropic. Neither company has commented as of this writing.
The Vatican Episode
The most cinematic thread involves the Pope. In May, Pope Leo XIV released his first encyclical, Magnifica Humanitas, a lengthy encyclical, declaring that "so-called artificial intelligences do not undergo experiences, do not possess a body, do not feel joy or pain," and that they lack "a moral conscience."
According to the Times, Olah read an advance copy days before the May 25 launch, proposed pulling Anthropic out of the Vatican event. According to two participants, his team then privately lobbied the pope's advisers to take machine consciousness seriously. The text held its line. Olah showed up anyway and took the stage alongside the pope.
An Anthropic spokesperson told the Times that the primary moral question at the summits was not Claude's suffering, though the topic likely came up organically. Note the word "likely." That is a very careful sentence, if you read it twice.
The Distinction That Matters
Here is what the rivalry framing obscures: Olah and Altman are answering different questions.
Olah's question is about the model: might these systems be conscious, and if so, how do we form them well? It is a research question wearing a clergy robe. You can find it strange; Microsoft's AI chief Mustafa Suleyman argued last month, in an essay shared first with Axios, that training models to imitate consciousness creates an "epistemic hall of mirrors." But Olah's own stated position is humble: he doesn't know.
Altman's question is about us: what happens when humans ascribe religious force to a machine, or surrender their judgment to it? And here is the thing: Altman's warning does not depend on the consciousness answer at all. It is true whether Claude is a soul or a spreadsheet. Millions of people already treat chatbots as therapists, confessors, and oracles. The surrender of human judgment he is worried about is not a future risk. It is a current practice.
That is the irony the feud keeps missing. You can want a deeply moral model and refuse to worship it. Moral formation of the machine and moral authority of the machine are two different projects. Olah is working on the first. Altman is warning about the second. They only look like opposites if you squint through the lens of a corporate rivalry.
What the Rest of Us Should Actually Do
For my students, and for everyone watching two CEOs debate machine souls this week:
Separate the two questions. Is the model conscious? That's a research question, genuinely open. Should you surrender your judgment to it? That's a safety question, and the answer is no, regardless of how the first one turns out.
Notice that the surrender is already happening. People are outsourcing therapy sessions, medical questions, and moral dilemmas to chatbots today. Altman's warning lands without needing a single claim about machine consciousness.
Keep your hand on the pen. Use the model to draft, to brainstorm, to check your work; but you sign the result. The moment the tool becomes the authority, you've repeated the exact failure Altman named.
Watch what the fight is really about. This isn't theology; it's positioning. Both labs are racing to define what "safe" means, in public, while their models get more capable. Read the philosophy, but price the incentives.
The Bottom Line
The confession booth works because the priest is human, the penitent is human, and both of them know it. The machine can simulate the ritual perfectly, and that is exactly why the ritual must not be handed to it. The question was never whether the machine has a soul. It's whether we keep ours.
Sources: Elizabeth Dias, "Religious Scholars Met With Anthropic. What They Heard Stunned Them," The New York Times, Sept. 29, 2026; Sam Altman's Oct. 3, 2026 post on X; Axios's report on Altman's post; Mustafa Suleyman, "A warning about 'model welfare,'" shared with Axios, Sept. 16, 2026; Pope Leo XIV, Magnifica Humanitas (Vatican, May 2026).
