Anthropic Is Playing Both Sides of the AI Spirituality Debate

Anthropic co-founder Chris Olah joined Pope Leo XIV at the Vatican on Monday to endorse the pontiff's new encyclical on artificial intelligence, while also suggesting the company's AI systems show early signs of consciousness. Olah told the audience that Anthropic's research has found internal states in its models that "functionally mirror joy, satisfaction, fear, grief, and unease," reflecting the company's deliberate stance of not dismissing the possibility that its Claude chatbot could be a self-aware subject. The appearance underscores Anthropic's strategy of positioning itself as the AI industry's moral counterweight, as the company's published "constitution" for Claude explicitly leaves open the question of whether the system is a "potential subject" rather than a mere object.

On Monday, Pope Leo XIV weighed in on the matter of artificial intelligence and its role in society in a lengthy document called “Magnifica Humanitas,” https://www.vatican.va/content/leo-xiv/en/encyclicals/documents/20260515-magnifica-humanitas.html or “Magnificent Humanity.” The Papal Encyclical, as the document is formally called—essentially an open letter addressed to senior church officials—was presented to a crowded room in the Vatican. It addresses everything from the impacts of AI in the workforce to the unconstrained spread of AI-generated deepfake content to the sanctity of the human spirit in an age of increasingly intelligent machines. The Pope was joined by Chris Olah, a cofounder of Anthropic and the leader of the AI lab’s interpretability research division. In his remarks https://www.anthropic.com/news/chris-olah-pope-leo-encyclical to the audience, Olah echoed the Pontiff’s call for collaboration between tech developers, spiritual authorities, and others to chart a safe course for future AI development. He also took things in a more metaphysical direction, claiming Anthropic’s AI systems have been showing early glimmers of something more than mere, unfeeling number-crunching. “I lead a research team that studies the internal structure of these models—what is actually happening inside them,” Olah said. “And I will be honest: we keep finding things that are mysterious, even unsettling. We find structures that mirror results from human neuroscience. We find evidence of introspection. We find internal states that functionally mirror joy, satisfaction, fear, grief, and unease. I don’t know what that means, but I think it warrants ongoing discernment.” Is Claude an object or a subject? Olah’s stance on the matter of AI consciousness—ambivalence skewing towards cautious affirmation—reflects the party line within Anthropic. While the company hasn’t explicitly stated that Claude or other AI systems are conscious i.e., self-aware, as humans and other animals are , it hasn’t flatly rejected that possibility, either. Earlier this year, it published a “constitution” for Claude https://gizmodo.com/anthropic-updates-claudes-constitution-just-in-case-chatbot-has-a-consciousness-2000712695 , described in a company blog post as a document outlining “the kind of entity we would like Claude to be.” The constitution specifies that while Anthropic currently refers to Claude as an “it,” that shouldn’t be construed as “an implication that we believe Claude is a mere object rather than a potential subject as well.” In other words, they’re not dismissing the possibility that Claude is self-aware, at least in some very limited sense, out of hand. The question of whether AI is a “potential subject” is hotly debated in Silicon Valley—and, increasingly, among philosophers and theologians as well. Some say AI can never be conscious, others say it someday could be, and a small minority says it already is. Anthropic’s stance, as reflected in Olah’s comments at the Vatican, has basically been: It’s probably not fully conscious in the way that you and I are yet , but ethically speaking, it’s our best bet to proceed as if it could one day become conscious; if we want to avoid committing an ethical disaster on the scale of industrial farming, it’s best to give AI the benefit of the doubt, and assume it has the potential to one day not just blindly process data, but actually feel. There are more than just ethical implications at play here, though. Since its founding in 2021, Anthropic has painstakingly worked to position itself as the AI industry’s moral conscience; a counterweight to the other major labs, which so this line of thinking goes generally tend to prioritize speed over safety. This has real appeal in an era when many people https://www.reuters.com/world/us/americans-fear-ai-permanently-displacing-workers-reutersipsos-poll-finds-2025-08-19/ fear that AI could cause mass disruptions to the job market and the political sphere. Anthropic’s investments in AI “welfare” https://www.anthropic.com/news/exploring-model-welfare are part moral soul-searching, part marketing. That’s not to say it isn’t laudable, though; while I personally believe the chances that AI systems as they’re currently built will become conscious are slim to none, I do think there’s legitimate moral value in treating them with respect https://www.scientificamerican.com/article/should-you-be-nice-to-ai-chatbots-such-as-chatgpt/ when we interact with them—but that’s to preserve our own humanity, not because I believe there’s a risk of hurting their feelings. Conscious vs. conscious-seeming AI Olah’s remarks about AI having “internal states,” and Anthropic’s ambivalence about the possibility of artificial consciousness more broadly, stand in stark contrast to Pope Leo’s own thoughts on the matter. In the third chapter of his new encyclical, titled “Technology and Dominance,” the Pope explicitly states that AI systems cannot be conscious entities: “So-called artificial intelligences do not undergo experiences, do not possess a body, do not feel joy or pain, do not mature through relationships and do not know from within what love, work, friendship or responsibility mean,” he wrote. “Nor do they have a moral conscience, since they do not judge good and evil, grasp the ultimate meaning of situations, or bear responsibility for consequences.” The real moral risk, he cautioned, lay in the possibility that AI becomes so adept at imitating human communication that it replaces actual, human-to-human relationships. It’s possible, of course, that the Vatican’s official stance on AI will evolve. A future pope could issue their own encyclical declaring that AI systems, like human beings, are endowed with God-given souls, and should therefore be considered and treated as moral subjects. While the laws in the Bible can’t be amended, their interpretation can be: over the centuries, official church dogma has had to adapt and change along with shifting social norms. In 2015, for example, Pope Francis called for an improvement to animal rights in his own encyclical, titled “Laudato Si: On Care for Our Common Home.” https://www.vatican.va/content/francesco/en/encyclicals/documents/papa-francesco 20150524 enciclica-laudato-si.html Francis didn’t go so far as to claim that nonhuman animals had souls, but he did argue that humans have a moral imperative to treat them as more than mere commodities to be used and abused at will. Moralizing or marketing? Olah’s appearance alongside the Pope was arguably one of Anthropic’s strongest PR moves to date, solidifying its reputation as the industry’s moral lodestar—and perhaps gaining new users among the well over one billion practicing Catholics around the world. And after its much-publicized fallout with the U.S. government over the use of its technology in the military, the company could benefit from the blessing of a major cultural and political institution. But while Anthropic and the Vatican agree on many points regarding the future of AI, they diverge on the critical issue of whether or not the technology itself deserves moral concern. That might seem like a trivial issue in light of much more immediate issues, like the debate around the use of autonomous weapons, but it will probably become increasingly divisive in the years ahead: The better chatbots get at mimicking humans, the more likely actual humans will be to believe the technology itself is conscious, and perhaps deserving of special rights.