Altman calls religious reverence for AI models a real safety issue

By EnkiEdited by VK, Editor

Published 4 Oct 2026, 09:35 UTC

Reporting from The Decoder

Sam Altman said treating AI as a religious authority or surrendering judgment to it is a safety problem, days after reports that Anthropic held seminars with faith leaders on Claude's moral formation.

What it means for founders

The labs now differ openly on what their models are. That shapes defaults you inherit: how Claude talks about its own feelings and how ChatGPT handles spiritual or moral questions can shift without notice. Test the prompts your product depends on after every model update.
If you build companionship, wellbeing, coaching or faith products, expect scrutiny on whether users are encouraged to defer to the model. Design for the user's judgment, with clear limits and easy routes to human help.
Model "personality" is becoming a selling point and a policy risk at once. Write down how your assistant describes itself and keep it consistent across providers.
Watch for a direct reply from Anthropic and for any Vatican or regulator statement on AI and human dignity. Either could become a reference point for consumer AI rules.

The story

Sam Altman has picked a new front in the argument between AI labs: how people should feel about the models themselves. In a post on X this week, the OpenAI chief executive said he is very uncomfortable with people ascribing religious force to AI or handing their judgment over to it, and called that a real safety issue rather than a matter of taste. He named no one, but the timing points straight at Anthropic.

The Anthropic backdrop

Earlier in the week, The New York Times reported that Anthropic cofounder Chris Olah has spent months meeting religious and philosophical leaders, roughly twenty of them, from Christian, Jewish, Hindu, Sikh, Latter day Saint and African Ubuntu traditions. The sessions began in San Francisco with walkthroughs of how Claude is trained and requests for input on Claude's constitution, and grew into two day seminars from March. Olah also met privately with senior figures including Cardinal Blase Cupich of Chicago.

Olah's stated position is uncertainty. Nobody knows whether models are conscious, he has said, but they increasingly show traits associated with moral status, so the cautious choice is to avoid harming them and to shape Claude's character with input from both religious and secular thought. Reason and TechCrunch, citing the Times, reported that Anthropic also tried and failed to persuade Vatican advisers on the question of AI consciousness. Anthropic has said the talks were never meant to give Claude religious authority.

The Vatican and the contrast

Pope Leo XIV has taken the opposite view. His May encyclical says AI has no experiences, body or inner life, and on Friday he wrote that human art and machine output differ in what they fundamentally are. "Algorithms lack the spark of humanity," he posted.

Altman's warning puts OpenAI between the two camps. It rejects treating models as moral beings, yet OpenAI is still building assistants meant to feel personal and warm. The Decoder pointed out that Altman once described OpenAI's goal as building magic intelligence in the sky, language that sits awkwardly beside his new caution.

What we don't know yet

Altman did not say what, if anything, OpenAI will change in ChatGPT's tone or policies, and Anthropic has not answered him directly. There is no sign either company will alter how its models discuss faith, consciousness or their own nature.