OpenAI CEO Sam Altman said on Saturday 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." The post on X on October 3, 2026 named no one — but it landed days after a New York Times report that Anthropic co-founder Chris Olah has been meeting with religious leaders about whether AI systems could be conscious.

Axios described Altman's post as a veiled shot at Anthropic. The Times reported that Olah met religious leaders to discuss whether models like Claude might have achieved consciousness, and how Anthropic could "inculcate morality" into increasingly powerful systems. Anthropic does not claim Claude is conscious, but its constitution for the model says the company is "not sure whether Claude is a moral patient" — and treats the question as serious enough to warrant caution.

The timing deepens the rift. One of OpenAI's own safety leaders, David Robinson, recently quit and publicly rebuked the company's safety record — just as the two labs, both approaching public offerings, are trading accusations over who handles AI risk responsibly. Altman did not explain what prompted the post, and neither company immediately responded to Axios's requests for comment.

Strip away the personalities and there is a genuine philosophical split here. Anthropic's constitutional approach treats the moral patienthood question — could Claude be the kind of thing that matters morally? — as serious enough to design around, even without an answer. Altman's position is that entertaining the question at all is dangerous, because it invites users to surrender judgment to the machine. Both positions are defensible; they point in opposite directions. One says caution demands taking consciousness seriously, the other says caution demands refusing to.

But the philosophy is also politics. With both labs approaching public offerings, the safety debate has become a competitive weapon — each side casting itself as the responsible adult and the other as reckless. Altman's "veiled shot" framing, Robinson's public resignation rebuke, the dueling narratives: this is what AI safety discourse looks like when billions in IPO valuation are on the line. The substance matters, but so does the timing, and the timing here is unmistakably strategic.

For everyone else, the practical question is narrower: how should you relate to these systems? Altman's warning has a plain-English core worth keeping regardless of the lab politics — do not outsource your judgment to a machine, no matter how fluent it sounds. Whether models are conscious is a question for philosophers; whether humans stay in charge of their own decisions is a question for all of us, and it is decided one uncritical delegation at a time.

What happens next is predictable: more public sparring, more manifestos, and eventually the IPO filings that force both companies to put their safety governance in writing for regulators. S-1 risk sections do not lie the way blog posts can. When Anthropic and OpenAI file, read those sections — they will tell you more about what each lab actually fears than any X post ever will.

Sources