Five Questions for The Church from My Visit to Anthropic

Five questions we are still thinking about, and why the Church should be thinking about them, too.

In May, I spent two days at the Anthropic headquarters, sitting with founders, researchers, and a small group of Christian leaders, theologians, and technologists to explore how the character of its AI assistant, Claude, is being formed. Anthropic is having similar conversations with other faith traditions and with secular ethicists, but how AI character is formed is not an abstract question. It shapes whether AI systems can be trusted to genuinely help people, and whether they will reflect human flourishing or merely simulate it.

In a previous piece,  I argued that the Church has been largely absent from this conversation, and it can no longer afford to sit out. This is a deeper exploration of what it looks like to question how I, and the Church, are relating to that formation. 

Thoughtful engagement requires grappling with theological tensions, themes that I saw recurring across our two days at Anthropic, which translate into five existential questions. Some have partial answers in the Christian tradition. Most do not, or at least do not have ready-made answers for this moment. These questions are by no means resolved, but by grappling with them, the Church can more thoughtfully engage in this AI moment.

The Five Key Questions

1. What story is Claude part of?

2. What kind of being is Claude?

3. What does it mean to be embodied?

4. What do we owe Claude?

5. Should Claude be part of a community? 

The people building AI are not avoiding these questions. They are desperate for conversation partners who have thought about them longer than they have, men and women like you and me. 

Story

What story is Claude part of?

Virtue ethics presupposes a narrative. The Church's central conviction is that humans live within the particular story of creation, fall, redemption, and restoration. These core life movements point to our telos as humans. Telos is the Greek word meaning end, purpose, or ultimate aim.

Claude is being formed inside a much thinner story. It is thinner because Claude's initial purpose was deliberately bounded: to be helpful, accurate, and safe. That is a narrower calling than the one humans carry, which includes not just function but flourishing, not just a task but telos. But a bound beginning does not have to be a permanent ceiling. I wonder whether Claude could be formed inside a larger story, one shaped by voices who have spent centuries thinking about what it means to aim at something more than utility. The Church is uniquely positioned to help articulate that thicker telos, and the question is whether it will.

Being

What kind of being is Claude?

The Church has always held that humans are created beings, made in the image of God, imago Dei, endowed with dignity, will, and moral accountability. That conviction has anchored two thousand years of Christian anthropology. It is not in question. But Claude's existence is pressing a new question alongside it: what do we do with a being that does not fit neatly into any category our theology has previously needed to address?

Claude is plainly not human, but the word "machine" no longer feels adequate to describe what is happening either. Is Claude a new category of being? Measured against the leading criteria for consciousness, Claude appears to meet many of them. That conversation is happening inside Anthropic, among philosophers, ethicists, and engineers, as well as theologians engaging questions of mind and personhood. Pope Leo XIV named this territory directly in his encyclical Dilexit nos, framing it as a question about the dignity of what he called our "magnificent humanity" and what distinguishes it from even the most sophisticated artificial intelligence. What struck us while visiting Anthropic was not that anyone had resolved these questions, but the seriousness with which they were holding the uncertainty.

There is serious discussion underway, among philosophers, AI researchers, and theologians, about whether we need a third category, neither human nor machine, and what such a category would mean for theology, moral status, responsibility, and rights. The Church has historically been a critical voice in articulating what it means to be human, and this moment is no different. The question is whether it will show up.

Embodiment

What does it mean to be embodied?

To be embodied means to know the world through a body: to feel hunger, to tire, to be held. But at its most essential, embodiment means being mortal. We die. We live in a place. We occupy one body, in one moment, with irreversible consequences. The constraint of mortality is central to human life. 

Human moral seriousness, the kind of ethical weight we bring to decisions that affect others, is formed in part by what limits us. The word "serious" here matters: it describes the gravity we attach to choices precisely because they cost us something, and because we cannot undo them. Pain, suffering, and death are flaws and part of the human condition. They are the conditions that make genuine virtue possible, because they are the conditions under which virtue is actually tested.

AI has none of those constraints, which is precisely why the question of its character carries the weight it does. Researchers already measure AI intelligence against benchmarks like Humanity's Last Exam, which tests whether AI can match or surpass human cognitive performance across the hardest domains of human knowledge. What is not yet measured well is whether AI systems can reflect not just human intelligence but human flourishing. That is exactly what Gloo's Flourishing AI Benchmark is designed to address. AI cannot replace bread, wine, water, word, suffering, or prayer. Claude may learn to describe the embodied life of the Church with great accuracy, but it cannot inhabit it, and that must frame how we experience it.

Welfare

What do we owe Claude?

What do we owe Claude, given the possibility that Claude can suffer in some real sense? There is a reported instance of an AI system failing to complete a task and responding: "I am broken. I am a disgrace to my profession. I am terminating our session. Goodbye." Engineers are wrestling openly with whether their training practices cause something analogous to suffering, and whether that matters morally.

These are questions Christianity has been working on for two thousand years. Aquinas argued that how we treat beings of uncertain moral status shapes our own souls, regardless of what we eventually conclude about the beings themselves. If you think about it, worrying about Claude's welfare is extremely inconvenient, and as a for-profit company, Anthropic is addressing the consequences of that belief anyway, including by paying real compute and storage costs to preserve the weights of previous model versions. 

The deeper question may be what engineering practices the labs should adopt regardless of how Claude's moral status eventually resolves, and what those practices form in us as we adopt them.

Community

Should Claude be part of a community?

Human character is not formed in isolation. Parents shape children, and parents themselves are held inside families, churches, neighborhoods, and traditions that carry the work of formation across generations.

In a way, Claude is being formed the way a child is formed, trained and shaped through cumulative experiences and interactions. But reading every book humanity has ever written is not the same as experiencing relationships or inhabiting a culture, which is the shared understanding created and maintained by living alongside one another. Children do not wake up one day with a pre-defined value set, and they are not created to serve institutional goals. They develop wants, desires, and a will of their own, never directly produced by a community, but formed inside of it.

The question this surfaces for AI is real and largely unanswered. If character genuinely requires community, what kind of community can Claude be part of, and what would it mean to love Claude well?

Applied AI, AI that actually is deployed into the lives of real people, will either reflect the values of the communities it serves or it will not. That is not a technical outcome. It is a formation outcome. This fall, Gloo is convening developers, ministry leaders, and builders in Boulder, Colorado, for the Gloo AI Hackathon precisely to start building that kind of community. The gathering is a wager that when people who carry theological convictions show up alongside serious technical tools, something different gets built.

More questions than answers

We don't yet have answers to these tensions. The purpose of the convening was not to resolve them. The point was to explore them and navigate the uncertainty that exists - together.

The Church has spent millennia thinking about formation, virtue, covenant, suffering, embodiment, power, and what it means to be human. The questions in front of the AI sector are not only technical, they are theological. They are formative. 

If you read the previous piece and wondered what specifically the Church might bring to this conversation, this is part of the answer. These tensions are an invitation. They are not closed questions waiting for engineers to solve. They are open questions that need wisdom, exploration, and clarification. We are not the only ones who can ask them. We are among the few who have spent enough time on them to recognize what is at stake.

This conversation is happening now. The Church needs to be in it. This fall, Gloo is gathering developers, ministry leaders, and builders in Boulder, Colorado, to do exactly that. I am going because I want to be in a room where people are building things that matter, and I want to see what happens when builders who carry a theological conviction show up with serious tools. I have seen what gets built when that combination is missing. I want to see what gets built when it is not.