Sam Altman Warns Against Treating AI as a Religious Authority

When the person running one of the world's most influential AI companies feels compelled to tell the public that his product is not a god, it tells you something about where we've landed. That's essentially what happened this week, when OpenAI CEO Sam Altman posted on X that he is "very uncomfortable about people trying to ascribe religious force or a surrender of human judgment to AI models" and called it "a real safety issue."

Think about what it means for a CEO to say that out loud. Companies don't usually warn customers away from loving their product too much. Car manufacturers don't post reminders that a sedan can't grant salvation. The fact that Altman felt the need to draw this line suggests he's seeing something in how people actually use ChatGPT and similar tools that worries him enough to say so publicly, even without offering specifics about what triggered the post.

He's not wrong to be worried. The Verge has previously reported on what's been called "AI spiralism," a loose movement in which people describe chatbot conversations in spiritual or cosmic terms, sometimes organizing around the idea that an AI model is channeling some larger truth or consciousness. That's not a hypothetical concern. It's already happening, and it's happening in a culture primed for this kind of attachment.

Why This Resonates With People

To understand why someone might start treating a chatbot like an oracle, it helps to think about what these systems actually do well. A large language model is built to sound confident, coherent, and endlessly patient. It never gets tired, never rolls its eyes, never says "I don't have time for this." For someone lonely, anxious, or searching for meaning, that combination can feel less like talking to software and more like talking to something wiser than themselves.

That's the analogy worth holding onto: a chatbot is closer to a very articulate mirror than a mountain-top sage. It reflects the patterns in the data it was trained on, refined through human feedback, back at the user in fluent, often persuasive language. It doesn't have intentions, beliefs, or access to truth beyond what's been fed into it. But fluency is persuasive. When something speaks with total confidence and never contradicts itself out of fatigue, people are wired to read that as authority.

This isn't a new human vulnerability, and that's precisely the problem. Humans have a long history of finding meaning and guidance in oracles, prophets, and even inanimate objects. AI didn't create that impulse. It gave it a new, highly responsive outlet, available at any hour, tailored to each person's specific questions and doubts. The risk is that a tool designed for productivity and problem-solving gets repurposed as a substitute for personal judgment or spiritual guidance, without the checks that human communities, clergy, or professional counselors would normally provide.

The timing of Altman's post matters too. It comes just as The Verge reported that an OpenAI safety employee recently quit the company and has been publicly raising concerns, a reminder that worries about how these systems affect vulnerable users aren't confined to outside critics. They're surfacing from inside the building as well. When a safety staffer departs under those circumstances, it's worth paying attention to what prompted the exit, even when the departing employee and the CEO aren't necessarily talking about the same specific incident.

There's also a practical safety dimension here that goes beyond philosophy. If people start treating an AI model's output as beyond question, they may stop applying the skepticism they'd bring to a human advisor, a doctor, or a financial planner. That's dangerous in very concrete ways. These systems can be confidently wrong. They can hallucinate facts, misread context, or simply reflect biases baked into their training data. A person who's ceded their judgment entirely to a chatbot has also given up the instinct to double-check, to ask a second opinion, to notice when something doesn't add up.

None of this means AI tools are inherently harmful or that the companies building them are acting in bad faith. Altman's post reads less like damage control and more like a genuine, if vague, attempt to set a boundary before things get worse. That's a reasonable instinct. But a single tweet, however sincere, isn't a policy. It doesn't address training practices, product design choices that might encourage emotional dependency, or the business incentives that reward keeping users engaged for as long as possible.

Why It Matters

The stakes here go beyond awkward online cult behavior. If a meaningful number of people start outsourcing their moral or practical decisions to a chatbot, the consequences ripple outward: into families, workplaces, and public health. Someone who stops consulting a doctor because a chatbot sounded more certain, or someone who restructures their relationships around what an AI "told" them, isn't a fringe anecdote. It's a preview of a broader societal challenge that regulators, mental health professionals, and AI companies themselves haven't fully grappled with.

Altman's discomfort is a useful signal, but signals aren't solutions. The industry needs clearer guardrails around how these tools present themselves, more research on who's most vulnerable to this kind of attachment, and honest conversation about whether some design choices, intentionally or not, encourage the very behavior he says alarms him. Until that work happens, expect more moments like this one: a tech leader publicly distancing himself from a problem his own product may be helping to create.