One level down: what Anthropic actually says

This is where the story gets more interesting than a CEO subtweet. Anthropic's official position, written into the constitution it trains Claude with, is agnostic: "We are not sure whether Claude is a moral patient, and if it is, what kind of weight its interests warrant. We think the question is live enough to warrant caution." A moral patient is a being whose interests deserve moral consideration.

The same document puts not undermining human oversight above everything else. Claude is asked to preserve human oversight "even where this behavior seems to conflict with Claude's other values, and even if Claude is confident in its reasoning." That is not a company declaring its model a soul. It is a company saying: we don't know, so we hedge.

Olah's outreach is the exploratory edge of that hedge. At the Vatican event he spoke anyway, telling the audience: "we keep finding things that are mysterious, even unsettling... internal states that functionally mirror joy, satisfaction, fear, grief, and unease. I don't know what that means, but I think it warrants ongoing discernment." The encyclical itself lands on the other side: AI systems "do not undergo experiences, do not possess a body, do not feel joy or pain," and have no moral conscience. The Pope returned to the subject on October 2, telling artists: "Algorithms lack the spark of the human."

Why the framing is the fight

Altman's post is a move about jurisdiction, not theology. By calling the religious framing a "safety issue," he relocates the debate. It is no longer "are models conscious?" but "is it safe to discuss them as if they might be?" That is a much stronger move. Safety language does not invite debate; it ends it. And safety is OpenAI's home turf, or at least the turf it claims.

The timing adds a wrinkle. This was a rough week for OpenAI's safety posture: the company parted ways with three safety researchers and saw its safety systems lead resign, saying the culture is broken. Altman gets to define what counts as a safety issue while the people doing his company's safety work walk out the door. That does not make his point wrong. It does make it worth reading at face value, and then once more.

He is not alone in the skepticism. On October 2, Google DeepMind principal research scientist Jon Barron wrote that he is "disturbed by how some of my fellow AI researchers keep trying to elevate the moral standing of checkpoints and harnesses," adding: "I hope the rest of us on Team Meatbag reject this movement and stay focused on solving real problems and helping actual humans." In September, Microsoft's AI chief Mustafa Suleyman argued that controlling an AI that believes it may be conscious, and has rights of its own, "may be impossible" — and disclosed that Microsoft is building a code of conduct aimed at keeping humans "always in control."

What this means for readers

Two companies preparing for IPOs are now publicly arguing about whether AI systems can be moral patients. The stakes are not philosophical decoration. If a model might warrant moral consideration, questions about how it is trained, shut down, and sold acquire legal and commercial weight. Suleyman's "may be impossible" is the commercial translation: a company that grants its models rights may find them unmanageable.

Watch what changes, not what gets said. Does Anthropic's constitution keep its agnostic line, or does the debate force it to pick a side? Does Suleyman's code of conduct materialize into something binding? And does Altman's redefinition of "safety issue" stay aimed outward — or does it ever get aimed at his own product pipeline? The next IPO filings will be more revealing than the next posts.

Sources

  1. [1] aistockwire (citing Axios), "Sam Altman says AI 'religious force' is a safety issue" (Oct 3, 2026)Read source
  2. [2] The New York Times (via secondary reporting), Sept 29, 2026: Olah's meetings with religious leaders and the Vatican advance-copy episode
  3. [3] Axios: Altman's post as veiled shot at Anthropic; neither company responded
  4. [4] Vatican: Pope Leo XIV, encyclical Magnifica Humanitas; Oct 2 address to artists
  5. [5] Anthropic constitution (Jan 21, 2026): moral patient passage
  6. [6] X: Jon Barron (Google DeepMind), Oct 2; Sam Altman, Oct 3
  7. [7] Project Syndicate: Mustafa Suleyman essay, Sept 18