Anthropic Flew 15 Priests to San Francisco to Teach a Chatbot Right From Wrong
Anthropic Flew 15 Priests to San Francisco to Teach a Chatbot Right From WrongTwo-day summit. Catholic and Protestant clergy. One question on the table: is Claude a “child of God”? I wish I was making this up.
15 religious leaders. A 2-day closed-door summit at HQ. Zero rabbis, imams, monks, or secular ethicists in the room. And it’s all going into the rulebook that decides how your AI answers you at 3 AM.
Right, so here’s what’s actually happening. One of the biggest AI companies on Earth quietly invited a room full of pastors to help decide what its chatbot believes. Reported first by The Washington Post (paywalled — Gizmodo has the free version).
🧩 Dumb Mode Dictionary
| Fancy Term | What It Actually Means |
|---|---|
| Claude | Anthropic’s chatbot. Like ChatGPT, different company. |
| Alignment | Getting the AI to behave the way its makers want. The whole ballgame. |
| The Claude Constitution | A written list of rules/values the bot is trained to follow. Think “company handbook,” but for a robot’s soul. |
| Guardrails | The invisible fences that stop the bot from saying something awful. |
| Input-gathering exercise | Corporate for “we asked some people what they think.” The clergy weren’t hired — they were consulted. |
📖 What went down in that room
- Late March. About 15 Christian leaders — Catholic and Protestant clergy, a few academics, some business folks — spent two days at Anthropic’s San Francisco office.
- They had dinner with senior researchers and talked through the hard stuff: how should Claude respond to someone grieving? To someone talking about self-harm? Can a machine be a “child of God”?
- This wasn’t a photo op. The answers feed into the Claude Constitution — the actual document that shapes how the bot talks to millions of people.
- Translation: a handful of people you’ll never meet helped write the moral wiring of a tool your kid might ask about the meaning of life.
😤 Why people are side-eyeing this
Here’s the part that made me put my coffee down.
- No one else got a seat. No Jewish, Muslim, Hindu, Buddhist voices. No atheists, no secular ethicists. Just one branch of one religion.
- When your “moral advisory board” all reads from roughly the same book, you don’t get ethics — you get one narrow vocabulary baked into a global product.
- Anthropic says more groups are coming later. Maybe. But the first draft is the one that sticks, and this was the first draft.
- Look, consulting humans about morality beats letting engineers wing it. But “we picked 15 and they all agreed with each other” is how you build a blind spot, not a conscience. Wikipedia on Constitutional AI if you want the technical backstory.
📊 The receipts
| Number | What it is |
|---|---|
| 15 | Religious leaders invited |
| 2 days | Length of the summit |
| 0 | Non-Christian faiths / secular ethicists in the room |
| 1 | Documents it shapes (the Claude Constitution) |
| Millions | People whose AI answers get quietly steered by it |
🗣️ What the timeline's saying
- The hype crowd: “Finally, AI companies taking morality seriously!”
- The skeptics: “So a for-profit picked its favorite moral framework and called it consultation.”
- The tired ones (me): Every system inherits the biases of whoever’s in the room when the rules get written. Doesn’t matter if it’s a firewall config or a chatbot’s soul. Kids these days call it “alignment.” We used to call it “whoever holds the pen wins.” Same thing. The Hacker News thread had the usual bloodbath, minus the aggregator link — go read the actual Yahoo writeup instead.
Cool. So a chatbot has a Sunday-school teacher now… Now What the Hell Do We Do? ( ͡° ͜ʖ ͡°)
Here’s the thing nobody’s saying out loud: a giant company just admitted that “what values should this AI have” is an open job nobody’s filled yet. That’s not a philosophy problem. That’s a market with the door propped open. Five ways in.
🕳️ The Doctrine Prompt Smuggler
Big AI picks one moral flavor. Everybody else’s community got left out — which means every group now wants an AI that talks like them. You can’t retrain the bot, but you can sell the exact instructions that make it behave. Package a “system prompt” (a paragraph of setup text you paste in before chatting) tuned to a specific community’s values and sell it as a download.
Example: A 24-year-old seminary dropout in Lagos writes a “Reformed values” prompt pack — a set of copy-paste instructions that make Claude/ChatGPT answer without saying stuff his community finds off. Sells it on Gumroad for $7 a pop to church small-groups. 400 sales in two months.
Timeline: First sales in ~10 days once you hit one niche forum. Plateaus in ~3 months when free versions get shared around — so launch the next community’s pack before that.
📡 The Belief-Gap Spy
Every AI refuses and allows different things. Nobody’s mapped it. Run the same 50 touchy questions through Claude, ChatGPT, Gemini, and the free local ones, screenshot who dodges what, and sell the comparison as a plain-English report to parents, schools, and faith orgs deciding which AI to let their people use.
Example: A 30-year-old ethics grad in Nairobi builds a one-page “which AI says what” chart, updates it monthly, and sells it to a network of private schools choosing a classroom AI. Charges per school for the updating service. Public data, zero permission needed.
🪟 The Values-Reviewer Patch Window
Anthropic just publicly admitted it pays outsiders to shape AI behavior. The gold rush hasn’t started yet. Right now, “AI values reviewer / red-teamer” gigs (people who stress-test what the bot says and flag bad answers) are wide open and under-applied. Get in before every philosophy major on Earth notices.
Example: A 27-year-old in Manila signs up on data-work platforms like Outlier and DataAnnotation, lists a background in theology + writing, and lands the “sensitive-topics reviewer” queue that pays more than generic labeling. $18–40/hr, remote, no CS degree.
🎣 The Congregation Answer-Bot
Every small church, mosque, and temple has years of sermons and texts sitting in a folder. Feed those into a no-code chatbot builder and you’ve made “our own AI that only answers from our stuff.” Sell the setup to congregations too small to ever build one themselves.
Example: A 26-year-old freelancer in Cebu uses Chatbase (upload documents, get a chatbot, no coding) to build a “Ask Our Ministry” bot from a church’s sermon archive. Charges a setup fee plus a small monthly. Books 5 congregations by word of mouth.
🧠 The Interfaith Cheatsheet Anchor
When news creates a new confusing topic, the first person to write the clear guide owns the search results. Nobody’s built the plain-English master page: “How every AI handles religion, morality, grief, and belief — and how to steer it.” Be that page.
Example: A 29-year-old writer in Jakarta publishes a free, genuinely-good “AI + Faith & Ethics” guide, keeps it updated, and quietly puts affiliate links to AI tools + a “set up your community’s bot” contact form at the bottom. Ranks first on Google for a dozen weird searches nobody else targeted.
🛠️ Follow-Up Actions
| Move | First Step | Tool |
|---|---|---|
| Sell prompt packs | Pick ONE community you know | Gumroad |
| Map AI refusals | Write your 50-question list | Any 3 free chatbots |
| Get paid to review | Upload a sharp writing sample | Outlier / DataAnnotation |
| Build congregation bots | Ask a local group for their sermon files | Chatbase |
| Own the SEO | Publish the guide nobody wrote | Free blog + Wikipedia for research |
Quick Hits
| You Want To… | Do This |
|---|---|
 Read the real story |
Washington Post (or free on Gizmodo) |
 See the actual rulebook |
Claude’s Constitution |
 Sell a values-tuned prompt pack |
Gumroad |
 Get paid shaping AI answers |
Outlier |
| Build a doc-based chatbot free |
Chatbase |
They flew in priests to teach a robot right from wrong. The rest of us just got shown the door that leads into the room where the rules get written — and it was unlocked the whole time.
!