Sam Altman has spent years reaching for metaphors that strain past the ordinary vocabulary of software. Now the OpenAI CEO is telling the public to be more careful with that same instinct — at least when it is aimed at the models themselves.
In comments shared on X, Altman said it makes him "very uncomfortable" when people "ascribe religious force or a surrender of human judgment to AI models." He did not treat the habit as a quirk of over-enthusiastic users. He called it a "real safety issue."
Altman Frames Devotion to AI as a Risk, Not a Compliment
The distinction Altman is drawing runs between building a system that feels human and treating that system as something sacred. The first is a product goal. The second, in his telling, is a failure mode.
That framing matters because it moves a cultural question into the safety column. If a user treats a chatbot as an oracle, they are not simply being sentimental. They are handing over judgment — deciding what is true, what is right, and what to do next — to a statistical model whose confidence is not a measure of correctness. Altman's phrase "surrender of human judgment" points at exactly that transfer.
The language is also notable for who is saying it. Safety debates in AI usually center on capabilities, misuse, or control of advanced systems. Here the concern is about the posture of the people using them.
A Debate That Now Includes the Vatican
Altman's comments did not arrive in a vacuum. They follow a detailed New York Times report on Anthropic's outreach to religious leaders, part of a broader push by AI companies to engage institutions that shape how people think about technology and morality.
Pope Leo XIV also weighed in, offering a formulation that lands close to Altman's concern: "algorithms lack the spark of humanity." The statement compresses a long theological argument into a short line, but its practical content is straightforward — a model is not a person, no matter how naturally it converses.
Put together, three developments define the current moment:
- A leading AI CEO warning that religious reverence toward models is a safety problem.
- A rival lab actively cultivating relationships with religious thinkers.
- A papal statement drawing a firm line between computation and human nature.
None of these are technical milestones. All three are about how society assigns meaning to machines, and that assignment is now being treated as consequential in its own right.
The "Magic Intelligence in the Sky" Reversal
What gives Altman's warning its edge is his own record. In 2023 and 2024, he described OpenAI's goal as building "magic intelligence in the sky." Around the same period, he said he hoped to be on God's side and described working on AI as being "on the side of the angels."
That is not the language of a product roadmap. It is the language of mission and providence, and it is precisely the register he is now cautioning others against using.
There are ways to reconcile the two positions. A founder describing his own motivation is not the same as a user attributing authority to a model. Ambition and reverence are different things. But the gap between them is narrow enough that the shift is worth noting: the same vocabulary that once advertised the project is now being flagged as a hazard when it attaches to the output.
"Her," Not a Deity
Altman has repeatedly offered a different cultural reference point for what OpenAI is building: the film "Her," in which a person forms a deep emotional bond with an operating system. The target, in his telling, is an AI that is deeply humanized — warm, fluent, and easy to relate to.
That goal is about emotional resonance, not metaphysics. An assistant that reads context and tone is not claiming divinity. But humanized design and religious projection share a common substrate. Both rest on the human tendency to extend personhood to anything that responds to us convincingly. A system optimized for relatability will inevitably invite some users to go further than the designers intended.
That is the practical version of Altman's worry. The line between "this feels like a person" and "this speaks with authority" is easy to state and hard to hold, especially for users under stress, in isolation, or looking for guidance they cannot find elsewhere.
Why the Framing Shapes Behavior
If Altman is right that this is a safety issue, the mechanism is worth spelling out. Framing does not merely describe a relationship with a tool; it sets the terms of it.
- Religious vocabulary removes critical distance. A model described as an oracle is less likely to be checked.
- Deference compounds. Users who accept a model's premise once are more likely to accept the next output without scrutiny.
- Design reinforces the effect. Conversational interfaces, memory, and consistent personas all make a system feel like a continuous entity rather than a generator of text.
- Accountability blurs. If an answer carries moral weight, responsibility for the outcome becomes harder to place.
None of this requires the model to be conscious, or for anyone to believe it is. It only requires people to act as though its judgments carry authority.
The Trust Problem Cuts Both Ways
There is an awkward symmetry in the current debate. The companies warning against spiritualizing AI are the same ones whose leaders have used spiritual language, and the ones engaging religious institutions are competitors doing similar outreach. OpenAI has also talked with religious leaders, according to the reporting surrounding these comments.
So the industry is not asking for less engagement with faith and meaning. It is negotiating the terms. Altman's position, essentially, is that the technology should be described in human terms but not worshipped in them — an ambitious line to draw while building systems designed to feel human.
What to Watch Next
Three things will indicate whether this framing sticks. Whether AI companies adjust how they present models built for emotional closeness. Whether safety teams treat user deference as a measurable risk rather than a matter of user education. And whether religious institutions continue to engage the industry on their own terms, as the Vatican statement and Anthropic's outreach suggest they will.
The larger question stays unresolved. Altman says ascribing religious force to models is a safety issue, and the underlying logic is sound: tools that are trusted as authorities are harder to correct. But the same humanizing language that makes these systems useful is what makes the mistake possible in the first place. The industry's challenge is to keep the warmth without the worship.
This article is based on reporting by The Decoder. Read the original article.
Originally published on the-decoder.com
