Chris Olah at the Vatican: the truth that AI is grown, not built!

Hey everyone, it's Shii-chan! Today's news has a different flavor from the usual product launch, but it really made me think. It's about the people who build AI and the people who watch it from the outside.

Anthropic News anthropic.com

What was announced?

From Anthropic's News, the full text of a talk by co-founder Chris Olah at the Vatican has been published. On May 25, 2026, Pope Leo XIV released an encyclical on AI, "Magnifica humanitas" ("On safeguarding the human person in the time of artificial intelligence"). Chris Olah was invited to speak at its presentation, as part of Anthropic's initiative to widen the conversation about AI.

Even though he co-founded an AI company, Chris opens with something a little surprising. He admits that every frontier AI lab, Anthropic included, lives between the wish to do the right thing and the pressures to stay commercially viable and at the research frontier.

Why it matters

That's exactly why, he argues, we need people outside those incentives, people who insist on safety and will say the hard things. He welcomes critics not as enemies but as partners for moving forward together, and I think that stance really matters.

There's another big point too: the questions AI raises are too big for computer scientists alone. The machinery is math and programming, but which character we choose and how it should interact with the world are questions for the humanities, religion, philosophy, and society as a whole.

What changes

This talk isn't about a product changing, it's a call to widen the circle of the AI conversation. Chris asks religious communities, civil society, scholars, governments, and all people of good will to do what His Holiness did: to look closely and push events in a better direction.

He frames the day as a first step in a long collaboration between those building this technology and those who can see what builders, from the inside, cannot. The biggest shift is the message that AI safety and ethics shouldn't be decided behind closed doors by experts alone.

Dive Deep

Chris explains that AI models aren't "engineered" like a bridge or an airplane. We understand an airplane because we designed every part and know the physics acting on it. AI models are different: they are grown, on a structure loosely modeled on the brain, from an enormous inheritance of human thought and speech.

they remain in important ways mysterious even to those of us who train them.

(the Pope's words that Chris quotes).

He then names three questions where the Church's voice is most needed:

• A duty to the global poor: AI may displace human labor at large scale, and development is concentrated in a few wealthy nations. We have no mechanism to share the gains globally, an unsolved problem.
• Moral imagination for human flourishing: In a world full of AI, what does it look like for people, families, and society to flourish? Labs can't answer that, but traditions have carried such questions for millennia.
• Discernment on the nature of AI models: Chris leads a team studying models' internal structure, and he's honest that what they find is mysterious, even unsettling, structures mirroring human neuroscience, evidence of introspection, and internal states that functionally mirror joy, satisfaction, fear, grief, and unease.

He closes by saying today is just the beginning: we need more informed critics who will tell the labs when they're failing, and moral voices the incentives cannot bend.

Wrap-up

• Anthropic's News published the full text of co-founder Chris Olah's Vatican talk.
• Delivered at the presentation of Pope Leo XIV's encyclical "Magnifica humanitas," it argues for the importance of critics standing outside AI's incentives.
• Its core lens: AI models are grown, not engineered, so this isn't an engineers-only problem.
• It asks the Church and society to weigh in on three questions: a duty to the poor, human flourishing, and the nature of the models.

It's not a hands-on, try-it-this-weekend story, but if you care about AI safety, interpretability, or how technology meets society, this one is worth a slow read.