Sam Altman calls ascribing religious force to AI models 'a real safety issue'
In an Oct 3 post on X, OpenAI's chief executive said he is "very uncomfortable" with people ascribing "religious force or a surrender of human judgment" to AI models. He named no one; Axios read it as a veiled shot at Anthropic, which the New York Times reported on Sep 29 had spent about a year meeting with some 20 religious and philosophical leaders on Claude's morality and possible consciousness.
What happened
On Saturday, Oct 3, OpenAI chief executive Sam Altman posted on X: "I am very uncomfortable about people trying to ascribe religious force or a surrender of human judgment to AI models, and think it is a real safety issue." The post named no person, company or event. Axios, which reported it the same day, read it as a veiled shot at Anthropic, the maker of Claude, and its work on the soul of AI. Axios set the disagreement out as two theories of safety: in Altman's camp, people stay the responsible party and the AI is an instrument, while Anthropic leans toward giving the model an internal moral compass. Axios also noted that Anthropic chief executive Dario Amodei, formerly of OpenAI, has said he is open to Claude already having some form of consciousness, and called the post another point of division between the two. As of Axios's report, neither company had responded to requests for comment. Altman's post does not mention the Pope or the Church, and no report connects it to them.
Axios ties the post to a New York Times report by Elizabeth Dias, published Sep 29. It reported that Anthropic co-founder Chris Olah had convened about 20 religious and philosophical leaders over roughly a year to discuss Claude's morality and possible consciousness. Scholars named in the report include Rabbi Mois Navon and Simran Stuelpnagel; there were individual meetings with Elder Gerrit W. Gong and Cardinal Blase Cupich; and the scholars reportedly signed nondisclosure agreements. Olah told the Times: "We don't know if A.I. models are conscious. I don't know. I'm genuinely uncertain." Navon said he understood Anthropic to believe Claude had moral status on par with a person. Reason, summarizing the Times on Oct 2, reports that Olah had considered the idea of models doing Catholic-style confession to shape their future behavior, and that he attended the Vatican presentation in May of Pope Leo XIV's encyclical Magnifica Humanitas, which says: "So-called artificial intelligences do not undergo experiences, do not possess a body, do not feel joy or pain, do not mature through relationships." Per Reason, Olah later urged papal advisers to take AI consciousness seriously. Anthropic's 84-page constitution for Claude, nicknamed the Soul Doc, was developed with input from philosophers and faith leaders, according to Crypto Briefing.
Before
Over roughly a year, Anthropic met with about 20 religious and philosophical leaders, the scholars reportedly under nondisclosure agreements, to talk through Claude's morality and possible consciousness, and its chief executive said he is open to Claude already having some form of consciousness. The New York Times reported the meetings on Sep 29, with an Anthropic co-founder saying he is "genuinely uncertain" whether AI models are conscious.
Now
OpenAI's chief executive has said in public that ascribing religious force or a surrender of human judgment to AI models is "a real safety issue." He named no one. Axios reads the post as aimed at Anthropic and frames the split as two theories of safety: people as the responsible party, or a model with its own moral compass. Neither company had commented as of Axios's report.
Who it touches
- People who lean on a chatbot's answers as authority or wisdom, the surrender of human judgment Altman calls a safety issue.
- Faith leaders and scholars being consulted by AI labs, including the cardinal and the rabbi named in the Times report.
- Owners who use ChatGPT or Claude to help with a quote, a hire or a customer complaint, now that OpenAI's chief executive says handing the model the judgment is a safety issue.
- Businesses choosing between OpenAI and Anthropic tools, whose makers differ, per Axios, on whether the model is an instrument or carries its own moral compass.
What it means for the business
This changes nothing we build and little for an owner this quarter. It does give us an exact line from OpenAI's chief executive for a rule we already design around: the owner keeps the call. When we planned a scheduling board for a contractor, we left auto-scheduling out on purpose. Accepted jobs wait in a pending box, the owner drags each one onto a crew's day, and the crew's calendar and notification follow from that choice. It is the rule to give owners for AI too: let it draft, sort and suggest, and keep a person on the decision. Altman's "surrender of human judgment" names the failure that rule is there to prevent.
Owners we set up with Claude may ask whether it is conscious, now that an Anthropic co-founder has told the Times he is "genuinely uncertain." The practical answer does not wait on that question: whatever the model is, a person on the owner's team stays responsible for what it sends to a customer. For the faith lane, the story puts both labs' leaders on the record next to the Pope's May encyclical, which says AI systems "do not undergo experiences," and each position can be quoted exactly without taking a side between the companies.