Altman Calls AI Worship a 'Real Safety Issue' as OpenAI Retreats From 'Magic Intelligence' Talk

Sam Altman said treating artificial intelligence as a higher power has become a genuine safety problem, a notable reversal from language he used about his own company's work as recently as 2024. The OpenAI chief executive warned that attributing religious or supernatural authority to AI models distorts how people use and trust them. His comments land as several AI labs and religious institutions are publicly circling the same questions about machine intelligence and faith.

The timing is not incidental. Anthropic has reportedly held meetings with religious thinkers, and Pope Leo XIV has issued a statement touching on AI, putting the technology's spiritual framing into mainstream debate. Altman's warning reads as an attempt to distance OpenAI from the quasi-religious expectations that have attached themselves to large language models. When users describe a chatbot as omniscient or infallible, they lower their guard against its errors — a failure mode that has real consequences when the system is advising on medical, legal or financial decisions.

Altman's own past remarks make the pivot sharper. In 2024 he described the goal as building what he called magic intelligence in the sky, and said he felt on the side of the angels doing the work. That framing helped recruit talent and investors during a period when OpenAI was selling a vision as much as a product. Now the same executive is arguing that such language, once it escapes the boardroom and reaches users, becomes a liability rather than a marketing asset.

The shift matters because AI companies have spent two years encouraging users to treat models as near-magical. Chatbots are marketed with language borrowed from science fiction, and some users have formed attachments that researchers compare to parasocial relationships. If the industry's most prominent CEO now says that posture is unsafe, it puts pressure on rivals to match the restraint — or to explain why they will not. It also gives regulators a new hook: if model makers themselves concede that mystical framing is dangerous, disclaimers and interface design become fair targets for rules.

What remains unclear is whether Altman's warning translates into anything concrete at OpenAI. The company has not detailed how it might discourage users from over-trusting its models, and its product roadmap still leans on capabilities that invite awe. The open question is whether this is a durable course correction or a rhetorical adjustment timed to a news cycle in which the Pope and a rival lab are already talking about AI and religion. Watch for whether OpenAI changes its own marketing, model behavior or user guidelines — or whether the language shifts again.