Sam Altman: Religious Framing of AI Is a Safety Issue

OpenAI CEO Sam Altman has spoken out against a growing habit of discussing AI models in religious terms. He says the trend makes him uneasy and that it creates real risk. His own history of describing AI makes the remarks more complicated than they first appear.

This edition looks at what Altman said, why the timing matters, and how his new position fits with years of his own sky-high rhetoric.

What Altman actually said

Altman said it makes him "very uncomfortable" when people "ascribe religious force or a surrender of human judgment to AI models." He described this as a "real safety issue."

That second phrase matters most. Altman is not objecting only to how AI is described. He is pointing to a specific failure mode: people treating a model's output as something closer to revelation than to a tool's suggestion, and giving up their own judgment as a result. In that framing, the danger is less about the model and more about how much authority users hand to it.

Why now: Anthropic, the Vatican and the Times

The comments did not appear in a vacuum. They followed two developments that pushed the topic into the mainstream.

The first was a detailed report in The New York Times on Anthropic's outreach to religious leaders. Anthropic, the company behind the Claude models and a direct OpenAI competitor, has been engaging with faith figures on questions about its AI systems. CyrioX recently covered how religious leaders described Anthropic's concerns about whether Claude could suffer.

The second was a statement from Pope Leo XIV, head of the Roman Catholic Church, who said that "algorithms lack the spark of humanity." For a global audience, that line carries weight. It came from one of the most widely followed religious authorities in the world, and it draws a clear boundary between software and people.

Read together, the two stories show that the line between AI development and spiritual questions has become a public debate. Altman's remarks place OpenAI on the side of keeping that line firm.

A record of sky-high language

The harder part is Altman's own track record. In 2023 and 2024, he said OpenAI's goal was to build "magic intelligence in the sky." Around the same time, he said he hoped to be on God's side and that he felt "on the side of the angels" working on AI.

OpenAI has also held conversations with religious leaders, so the outreach Altman now seems wary of is not unique to Anthropic.

None of this proves that Altman ever claimed AI models had a soul or deserved worship. But "magic intelligence in the sky" is not neutral engineering language. It invites exactly the kind of awe that he now calls a safety concern. Readers can reasonably ask whether the new caution reflects a change in thinking or a change in messaging.

The "Her" ambition

There is another thread worth noting. Altman has repeatedly said his goal is AI on the level of "Her," the science fiction film in which a man forms a deep emotional bond with an AI system. The aim, in other words, is a deeply humanized assistant.

That is not the same as giving models consciousness or religious meaning. A product can feel human without anyone claiming it is divine. Still, the two ideas sit close together. A system designed to feel like a trusted companion is also one that people may be inclined to defer to. Whether spiritual talk about building humanlike AI is more trustworthy than religious talk about the models themselves is a judgment each reader will have to make.

Our Take

The core point Altman makes is sound. When people stop questioning what a model tells them, mistakes and manipulation become harder to catch. That concern sits alongside other recent debates over how AI companies frame their technology, including a DeepMind essay arguing for symbiotic AI rather than a singularity narrative. This suggests at least part of the industry is moving away from grand, near-mystical language.

The tension is that the same companies have benefited from that language. Talk of magic and angels helped build excitement, and humanlike products encourage the trust Altman now warns about. It is worth watching whether OpenAI's product design and marketing follow his words, or whether the caution remains a talking point aimed at a rival. Another useful signal will be whether religious institutions keep shaping how AI companies describe what they are building.