On May 15, 2026, Anthropic co-founder Chris Olah stood at the Vatican’s Synod Hall and delivered a speech that most tech coverage missed entirely. The event was the official launch of Pope Leo XIV’s first encyclical, Magnifica Humanitas. The headlines focused on the Pope. They should have also focused on Olah.
What he said in six minutes and forty-three seconds is one of the more honest public statements to come out of the AI industry in years.
He didn’t come to represent the industry. He came to ask for help.
Olah opened by acknowledging something unusual coming from someone in his position: every frontier AI lab, including Anthropic, operates inside incentives that can conflict with doing the right thing. Commercial pressure, geopolitical pressure, pride, ambition. And that no matter how sincerely anyone intends to do the right thing, those incentives will always have influence.
His ask was direct: they need people outside those incentives. People who care about things going well, who are paying close attention, who are willing to say hard things, insist on safety, and be earnest critics. Not consultants. Not regulators. People whose judgment the money cannot bend.
The Church didn’t seek out Silicon Valley. One of the most powerful AI companies in the world walked into the Synod Hall and said: we need you to hold us accountable.
Why AI is not an engineering problem
Olah made a distinction that rarely surfaces in mainstream AI coverage. AI models are not engineered the way a bridge or an airplane is engineered. We understand an airplane because we designed every part of it and we understand the physics acting on it. AI models don’t work that way.
They are grown — his word — on a structure roughly modeled after the brain, trained on an enormous inheritance of human thought and speech. The result is something far more subtle and strange than the cold, calculating robots science fiction prepared us for. And it remains, in important ways, mysterious even to the people who build it.
That’s not a talking point. Olah leads the research team at Anthropic that studies the internal structure of these models — what is actually happening inside them. And he said, plainly, that they keep finding things that are unsettling. Structures that mirror results from human neuroscience. Evidence of introspection. Internal states that functionally resemble joy, satisfaction, fear, grief, and unease.
He doesn’t claim to know what that means. But he thinks it warrants serious, ongoing discernment — and that computer scientists alone are not equipped to provide it.
Three problems he says the industry can’t solve alone
Olah named three questions where he believes outside voices are especially needed. The first is the displacement of labor at large scale — and more specifically, the fact that AI development is concentrated in a handful of wealthy nations with no existing mechanism to share the gains globally. An unsolved problem, he said, of the kind that institutions like the Church have historically refused to let the world ignore.
The second is moral imagination: what does it actually look like for humans, families, and communities to flourish alongside AI? That’s not a question a lab can answer. It’s a question that religious and philosophical traditions have been carrying for millennia, and he asked them to keep carrying it into this moment.
The third loops back to the mystery inside the models — the internal states, the structures that mirror the brain, the things that keep surprising his own research team. He wants more discernment on what these systems actually are, not just what they do.
They are made from us, from our words
Near the end of his speech, Olah said something that doesn’t sound like a tech executive and doesn’t sound like a press release. Describing what AI models actually are — what they’re built from — he said this: they are grown on an enormous inheritance of human thought and speech. They are not the robots we were promised. And then, simply: “They are made from us, from our words.”
That framing changes the conversation. If these systems are built from the accumulated record of human expression — our literature, our arguments, our questions, our fears — then questions about their nature are not purely technical. They are questions about what we put into them. And the people best equipped to think about that may not be the ones writing the code.
That’s what Olah went to the Vatican to say. And it may be the most important thing anyone in the AI industry has said in public this year.
There is something else sitting underneath all of this, harder to name. Humanity has spent decades — centuries, really — imagining this moment. In books, in films, in philosophy, in nightmares. We rehearsed it endlessly, from every angle. And now that it’s here, most people are going about their day. Some sense that something has shifted without being able to say exactly what. And a few — Olah apparently among them — have looked directly at it and felt the ground move. What’s strange is not that we built this. What’s strange is that we always knew we would, and we’re still not sure we were ready.
This article is part of a series on AI, culture, and power. Also read: When “Disarm AI” Actually Makes Sense and We Imagined It for Two Centuries. It Still Caught Us Off Guard.
