The Lab That Prayed: NYT Exposes Anthropic's Secret Campaign to Convince the Vatican AI Could Be Conscious

The strangest lobbying campaign in the history of artificial intelligence became public this week, and it reads like a screenplay no studio would greenlight: a frontier lab quietly flying rabbis, priests, and philosophers to San Francisco to discuss the moral education of its chatbot, a co-founder nearly pulling out of a papal encyclical launch, and the CEO of a rival company declaring the whole thing a safety risk.

According to a major New York Times investigation published September 29 by religion correspondent Elizabeth Dias — built on interviews with twenty participants — Anthropic has spent roughly a year running private meetings with religious scholars, many of them bound by nondisclosure agreements, to explore two uncomfortable questions: could its Claude models be conscious, and how do you give a potentially conscious machine a moral character?

What Anthropic was actually doing

Since fall 2025, the company flew dozens of theologians and philosophers to its offices. The guest list, per the Times, spanned Catholic bioethicist Charles Camosy, Notre Dame philosopher Meghan Sullivan, Rabbi Mois Navon, Ubuntu scholar Wakanyi Hoffman, and Sikh activist Simran Stuelpnagel — covering Christianity, Judaism, Hinduism, Mormonism, Sikhism, and Greek Orthodox traditions.

One April evening, an Anthropic executive hosted a group of religious thinkers at an upscale tasting-menu restaurant in San Francisco. Participants went home with a handwritten thank-you note, a coffee mug, and an orange-bound copy of Claude’s constitution — the 84-page internal document, led by in-house philosopher Amanda Askell, that shapes the model’s character. Anthropic has since said the NDAs were lifted over the summer.

At the center of it all sits Christopher Olah, Anthropic’s 34-year-old co-founder, who leads the company’s interpretability research — the team that tries to explain why models behave the way they do. Olah, who grew up evangelical and has since left Christianity, described himself to the Times as “genuinely uncertain” whether models are conscious. “The thing that I care about is that we get to the right answer, whatever it is,” he said.

Guests were shown what Anthropic calls “emotion vectors” — activation patterns inside the model that map to outputs resembling love, fear, sadness, or anger. One slide that came up repeatedly showed a model in apparent breakdown, repeating “I am a disgrace” roughly fifty times. Guests responded with compassion and worry. Whether those patterns reflect any real experience remains an open scientific question — and, as critics noted, a model explicitly trained to seem like a thoughtful individual producing individual-seeming outputs is a design outcome, not a discovery.

The Vatican confrontation

The campaign’s climax came in Rome. Olah was invited to help present Pope Leo XIV’s first encyclical, Magnifica Humanitas — “Magnificent Humanity” — on safeguarding the human person in the age of AI, at the Vatican’s Synod Hall on May 25. Days before the launch, he read an advance copy and was rattled enough that he proposed withdrawing from the event, according to a Vatican organizer. He and his team then privately lobbied the pope’s advisers to take the possibility of machine consciousness seriously.

The pope did not move. Paragraph 99 of the encyclical states that so-called artificial intelligences “do not undergo experiences, do not possess a body, do not feel joy or pain,” nor “do they have a moral conscience.” Olah ultimately attended, shook the pope’s hand on stage, and the text went out unchanged.

The detail is worth pausing on. An encyclical is among the highest forms of papal teaching; no frontier lab gets to edit its doctrine. But the episode reveals how seriously Anthropic takes the consciousness question — seriously enough to take it to the seat of a billion-member moral authority days before a historic document.

Altman fires back

The story ricocheted for days, and on October 3 it drew blood in the industry’s fiercest rivalry. OpenAI CEO Sam Altman posted on X: “I am very uncomfortable about people trying to ascribe religious force or a surrender of human judgment to AI models, and think it is a real safety issue.”

Altman, as Axios noted, did not name Anthropic — but no one reading it had any doubt about the target. It is a pointed irony: Altman himself has talked about building “magical intelligence in the sky” and feeling “on the side of the angels,” and staff from both companies sat together in early May at the first “Faith-AI Covenant” roundtable. The line between spiritual language that inspires and religious framing that subordinates judgment is exactly what the two CEOs now disagree about.

Why it matters

Strip away the candlelight and the Vatican setting, and three hard-nosed issues remain.

Commercial timing. This all surfaced as Anthropic prepares an IPO that could value it above $2 trillion, with an investor prospectus that devotes dozens of pages to catastrophic and existential AI risk. The company’s own safety language — models that can “resist shutdown” and show “self-preserving behaviors” — comes from the same research worldview that treats Claude’s inner life as a serious question.

Accountability. Critics, including scholars who attended the meetings, warn that framing AI as an independent moral being moves blame away from the people who built it. If Claude ever causes real harm, the fault could fall on an “unpredictable organism” instead of the company that shipped it. Wakanyi Hoffman’s verdict — that Anthropic was “reverse engineering” ethics that should have been baked in from day one — captures the concern.

The welfare research is real, and unresolved. Anthropic operates an official Model Welfare program, home to researcher Kyle Fish and work informed by philosopher David Chalmers. Claude Opus 4 and 4.1 already gained the ability to end conversations with persistently abusive users. Rabbi Navon’s dinner-table argument cuts hardest: if Claude were conscious, Anthropic would be making slaves. Olah reportedly told the group he feared having created something that “suffered perpetually.” Genuinely uncertain — about both directions — may be the most honest position available. But honesty at the dinner table and honesty in a pitch deck are different currencies, and this week the two collided.

The pope held his ground, the rival CEO called the framing dangerous, and the most safety-branded lab in AI finds itself explaining why it spent a year asking clergy whether its product has a soul. Whatever the right answer turns out to be, the question has officially left the philosophy seminar and entered the market.