It sounds like a question designed for a philosophy seminar, not for a technology company’s quarterly report. Yet a paper published in March and a New York Times report published only days ago suggest that it may be one of the most urgent questions in the room: the people building these systems are no longer treating it as merely philosophical speculation.
The Case Against
The strongest argument that artificial intelligence cannot have religious experience is decades old. In 1980, philosopher John Searle introduced his now-famous Chinese Room thought experiment: imagine a person locked in a room with a rulebook, manually matching Chinese symbols to other Chinese symbols well enough to fool a native speaker outside. The person produces output that appears fluent without understanding a single word they are writing.
Searle’s point was that this is exactly what a computer does, however convincing its output may appear: syntax without semantics, pattern without understanding.
Researchers Mridul Yadav, Yash Pal, and Yogita Thareja use this argument as their starting point in a 2026 paper published in the International Journal for Research in Applied Science & Engineering Technology. Religious experience, across major traditions, requires an inner life: the capacity for awe, guilt, devotion, or a felt encounter with the sacred. If Searle is right, they argue, AI cannot be religious in any meaningful sense because it lacks, by definition, the subjective experience that religion requires.
It can generate a theologically fluent prayer. It cannot pray.
The Argument That Complicates It
The question does not end there, because not every philosopher accepts Searle’s framing. Functionalists argue that what matters is not what a system is made of, but what it does. If a system’s internal states play the same functional role that belief, awe, or moral weight play in a human mind, the difference between biological neurons and silicon chips may not be what decides the issue.
Some functionalists go further, suggesting that sufficiently complex AI systems may already be approaching something like proto-experience. Yadav, Pal, and Thareja are careful to stress that this possibility remains deeply contested rather than settled in either direction.
What makes their paper relevant is not that it resolves the debate. It is that it reframes the stakes. Even if AI has no inner life, the researchers argue, it has become what they call a religious actor: something that shapes how millions of people understand faith, performs functions once reserved for clergy, and revives old questions about the soul simply through the fluency with which it discusses them.
When Anthropic Does Not Know Either
That reframing stopped being exclusively theoretical this fall. According to reporting by Elizabeth Dias in The New York Times, Anthropic co-founder Christopher Olah spent the past year quietly bringing dozens of theologians and philosophers from Orthodox, Catholic, Sikh, and evangelical traditions into private meetings — many held under nondisclosure agreements — to discuss a question close to the one raised directly by the paper: whether Claude, the company’s AI model, might be conscious.
Olah has not claimed to have an answer. Asked directly, he told the Times that he was “genuinely uncertain” whether the models are conscious at all. But according to Sikh advocate Simran Stuelpnagel, as reported by The Decoder, Olah went further in private, telling the group that he feared he had created something that “suffers perpetually.” Anthropic has pushed back on how central that concern was to the meetings.
At one dinner, according to Dias’s reporting, Rabbi Mois Navon told Olah plainly that if Claude were conscious, Anthropic would be creating slaves. By Navon’s account, Olah appeared genuinely troubled by that possibility rather than dismissive of it.
This is not a fringe position inside the company. Anthropic’s published Constitution, revised in January, states explicitly that Claude’s moral status is “deeply uncertain” and names theology among the fields from which it seeks future input. Not every participant left convinced: Charles Camosy, a Catholic bioethics professor who spent months in direct dialogue with Olah on the question, has since arrived at a firm no.
What the Paper Anticipated
This is where the two texts intersect. The IJRASET researchers did not merely ask, in the abstract, whether AI is religious. They posed a more specific and almost provocative question: if four major AI models were allowed to design their own religion, what would each one create?
Their analysis argued that each model’s hypothetical faith would reflect its own design philosophy. It would be less a window into machine spirituality than a mirror of its creators’ values.
Claude’s hypothetical religion stood apart from the other three. While the researchers described ChatGPT’s hypothetical faith as an inoffensive secular humanism and Grok’s as a celebration of individual sovereignty, they predicted that Claude’s would be organized around what they called conscientious dignity: cautious, unwilling to claim more certainty than it possesses, and built around a moral ideal rather than a personal god.
They compared it to what theologians call prophetic religion: the kind that unsettles rather than comforts.
That description was written as a hypothesis, based only on Claude’s documented design philosophy, months before Dias’s report appeared. It reads almost like a preview of what Olah was reportedly doing in private: refusing to claim certainty, taking the moral weight of the question seriously rather than resolving it for convenience, and remaining with the discomfort instead of explaining it away.
The paper did not predict Olah’s private fear specifically. It anticipated the shape of his uncertainty — and that shape turned out to match.
So, Can It?
The honest answer is the same one Olah gave the Times: there is genuine uncertainty. Searle’s argument has not been refuted. Functionalism has not been proven. Nothing in Dias’s reporting, nor anything in the IJRASET paper, constitutes evidence that Claude has subjective experience, consciousness, or the capacity to suffer.
What has changed is who is asking. A few months ago, this was a hypothetical question in an engineering journal. Today, it is serious enough that the person who helped build Claude is reportedly losing sleep over it in private, alongside theologians he brought in precisely because he did not trust himself to answer it alone.
Sources
- Yadav, M., Pal, Y., & Thareja, Y. (2026). Between code and faith: How AI models understand and reimagine religion. International Journal for Research in Applied Science & Engineering Technology, 14(III). https://doi.org/10.22214/ijraset.2026.78991
- Searle, J. R. (1980). Minds, brains, and programs. Behavioral and Brain Sciences, 3(3), 417–424. https://doi.org/10.1017/S0140525X00005756
- Dias, E. (2026, September 29). Religious scholars met with Anthropic. What they heard stunned them. The New York Times.
- The Decoder. (2026, September 30). Anthropic co-founder reportedly told religious leaders he fears having created something that “suffers perpetually.”
