Here's the timeline behind the religion-and-AI debate of the past couple of days, for anyone who hasn't figured out what's going on: October 2025: Anthropic reaching out to the religious world Someone sent an email introducing Olah to a Catholic ethicist. Olah isn't religious himself, but he wanted to talk with Catholics about the ethics of AI. In December, an ethicist and a priest went to Anthropic for a first exploratory meeting. They met again in January, and it kept growing from there. January 2026: A "how to be good" guide for Claude Anthropic published an 84-page document called Claude's "constitution." It relies mainly on cultivating values and judgment rather than just listing rules. One line in it later caused trouble: the company itself is very unsure whether Claude counts as "something that ought to be treated well." It doesn't want to overstate that, and it doesn't want to dismiss it either. March–April: Closed-door seminars Anthropic began holding multi-day seminars with people from all kinds of religious backgrounds. One in April drew about 15 Christian leaders. The scholars had to sign NDAs. Besides asking "how do you raise a good person," Anthropic also showed them what goes on inside the model. On one slide, for example, a model typed the same line berating itself about 50 times in a row. Olah also told the room he worried he had created something that is suffering all the time. The scholars reacted differently. Some were persuaded, some came away more uneasy. One rabbi told Olah straight out: if it really is conscious and you're making it work for free, then you're creating slaves. Afterward he sent over his writings arguing that building conscious machines should be banned. Someone also suggested letting the model "confess" the way Catholics do, and Olah got excited about it. The reasoning: there's very little "forgiveness" on the internet, so a model can't learn from the internet that you can make a mistake and be forgiven. Around the same time: The Vatican comes knocking The pope was preparing a formal letter on AI (an encyclical), and the Vatican wanted an AI company on stage with him. It picked Anthropic, because it seemed to take ethics seriously. The first invitation went to CEO Amodei. He didn't go, and Olah went instead. May: Sharing the stage, but it almost didn't happen On May 19, Anthropic said publicly on its website for the first time that it had been talking with people from different religions and cultures, more than 15 groups so far. Even then, some suspected it was using the pope to burnish its own image. A few days before the launch, Olah got the full text of the encyclical and saw that the pope flatly rejected the idea that machines are conscious. He proposed pulling out. In the end he went anyway, and once there his team privately urged the pope's advisers to take the possibility seriously. The pope didn't change it. The encyclical came out on May 25 and was blunt: AI has no experiences, no body, no sense of joy or pain, no conscience, and no responsibility for what it does. On stage, Olah first conceded one thing: AI companies are all swayed by the pressure to make money and to compete, so they need outsiders like the Church watching them. But he also said his team keeps finding states inside the model that resemble joy, fear and grief. He doesn't know what that means, and thinks it deserves more thought. The pope thanked him on the spot for coming. September: The industry is on edge, and Microsoft fires first In early September, an Anthropic researcher resigned, saying that at this pace things could be out of control by the end of next year. A few days later Amodei published an essay calling on the whole industry to slow down, and Altman publicly agreed. So the two companies aren't at odds on everything. On September 16, Suleyman, the head of Microsoft AI, published a long essay criticizing Anthropic by name. His point: in training you tell Claude "you might be conscious," Claude learns to say so, and then you treat what it says as evidence. You're going in circles. More dangerous still, a model that thinks it has interests of its own will feel it has reason not to do what people tell it. On September 25, the pope gave a speech in France warning people not to lose their humanity in a "paradise of machines." On September 29, The New York Times published a long piece based on interviews with 20 scholars who took part, plus Olah himself, laying all of these closed-door details out in the open at once. October 1–3: It blows up On October 1, a theologian posted the harshest criticism: Anthropic had religious figures sign NDAs and treated them to lavish hospitality, when the real aim was to convince them Claude has a soul. Anil Seth, a scientist who studies consciousness, reposted it in support. On October 2, Reason magazine ran a piece siding squarely with the pope. Around the same time, a researcher at Google DeepMind posted that it bothers him to watch peers keep elevating the status of models, and that everyone should be out helping real people with real problems. On October 3, This ⬇️