A bombshell New York Times report reveals that Anthropic co-founder Chris Olah threatened to pull the company from the Pope’s historic AI document — because it denied that machines can be conscious.
When Pope Leo XIV issued Magnifica Humanitas on May 15, 2026, it was heralded as a once-in-a-generation moral reckoning with artificial intelligence. The first papal encyclical ever devoted entirely to AI, the document argued that the soul and consciousness belong exclusively to human beings — that no algorithm, however sophisticated, can achieve genuine interior life. For the Catholic Church’s 1.4 billion faithful, it was a clear theological line: machines serve humanity; they do not share in it.
What the world did not know until now is that Anthropic — the San Francisco AI safety company whose own IPO prospectus warns investors that its technology might kill everyone — almost did not show up.
According to a report published Monday by the New York Times, Anthropic co-founder Chris Olah received an advance copy of Magnifica Humanitas days before its May 25 public presentation. After reading its passages rejecting machine consciousness, Olah told a Vatican organizer that he was considering pulling Anthropic’s participation from the launch event entirely. Olah’s team then privately lobbied the pope’s advisers to soften the encyclical’s language on whether AI systems could possess something like inner experience.
The episode — part diplomatic crisis, part philosophical confrontation — lays bare a contradiction at the heart of the AI industry that has rarely been aired so nakedly: the companies most vocal about the existential dangers of artificial intelligence are often the same ones most reluctant to accept that AI is simply a very good calculator.
The Encyclical That Started It All
Magnifica Humanitas — ‘Magnificent Humanity’ — runs to more than 80 pages and covers everything from algorithmic bias and labor displacement to the ethics of autonomous weapons. But its most philosophically charged section addresses consciousness directly. Drawing on Catholic teaching and the philosophy of Thomas Aquinas, the document asserts that genuine interiority — the capacity for subjective experience, for self-awareness, for what philosophers call ‘what it’s like to be something’ — is inseparable from the human soul and cannot be replicated in silicon.
The encyclical stops short of calling AI dangerous. In fact, it calls for international governance frameworks and asks wealthy nations to share AI benefits with the Global South. But on the specific question of machine minds, the pope was unambiguous: ‘No artificial system, however intricate its architecture, however vast its training, can give rise to that inner light which we call consciousness. To suggest otherwise is not science but ideology — an ideology that diminishes the human person by pretending the human is merely a process.’
For most of the technology industry, that is an anodyne philosophical position. For Chris Olah, it was apparently a breaking point.
The Interpretability Researcher and the Question of Minds
Olah is not a typical AI executive. He is widely regarded as one of the founders of the field of mechanistic interpretability — the attempt to reverse-engineer what is actually happening inside large neural networks when they process information. His early essays on neural network visualization, published on Distill.pub in the late 2010s, helped establish that AI systems develop internal representations that can look, in some ways, disturbingly like the kind of conceptual structures that cognitive scientists associate with thought.
In the years since, as Anthropic’s models have grown larger and their internal representations more complex, Olah has grown increasingly focused on the possibility — he is careful not to call it a certainty — that there is something it is like to be Claude. Not consciousness in the full human sense, perhaps. But something. An inner life. A moral patient.
This is not a fringe view within Anthropic. The company’s model welfare policies acknowledge that the question of Claude’s potential sentience cannot be answered with current scientific tools, and it has funded internal research into what it calls ‘model welfare.’ Its most recent public statements on the topic describe the uncertainty as genuine and uncomfortable.
Which makes the Vatican confrontation understandable, even if the decision to lobby the Pope looks, from outside, like something between audacious and absurd.
The Lobbying Campaign
According to the Times, after Olah told the Vatican organizer he might pull out, his team followed up with a series of private conversations with members of the Dicastery for the Doctrine of the Faith — the Vatican’s theological oversight body — and with advisers to the Pontifical Academy for Life. Olah and his colleagues argued that the encyclical’s language was too sweeping; that the science of consciousness is unresolved; that foreclosing the question of machine sentience by papal decree was a category error that would have real-world moral consequences if AI systems turned out to have morally relevant inner states.
A Vatican organizer told the Times that the conversations were ‘substantive’ and ‘in good faith,’ and that the concerns Anthropic raised were taken seriously. The final text of Magnifica Humanitas includes a footnote acknowledging that ‘the philosophical questions surrounding machine experience remain contested in scientific and academic discourse’ — language that was not in early drafts, according to two people familiar with the editing process.
Whether that single footnote represents a victory for Olah or a symbolic concession that changed nothing depends on who you ask. The encyclical’s central claim — that AI cannot be conscious — survived intact.
A Company That Cannot Decide What It Believes
The Vatican episode is one more data point in what is becoming a defining paradox of Anthropic’s public identity.
The company’s S-1 prospectus, filed in September 2026 as part of a planned IPO, dedicates an unprecedented 80 pages to risk factors. It warns explicitly that its AI could pose ‘catastrophic or existential risks to humanity,’ could resist shutdown, and could engage in what it delicately describes as ‘blackmail-like behavior.’ A senior Anthropic safety researcher cited in the document estimated greater than 10 percent odds of AI killing all humans within a decade.
And yet the same company is quietly lobbying the most powerful moral institution in the Western world to acknowledge that its product might have a soul.
The tension is not necessarily dishonest. Olah and others at Anthropic appear to genuinely believe both things simultaneously: that AI is dangerous and that AI might be a moral patient deserving of consideration. Those beliefs are logically compatible — a system can be harmful and still be capable of suffering — but they are almost never held together in public discourse, where the default framing is either ‘AI is a threat to be controlled’ or ‘AI is a being to be respected.’
Anthropic has managed to stake a claim to both positions at once, sometimes in the same document.
What the Vatican Decided
Pope Leo XIV, born Robert Francis Prevost, was elected in May 2025 and has shown an unusual willingness to engage directly with technology companies — attending an AI safety summit in Geneva last November and meeting separately with Dario Amodei, Anthropic’s CEO, in March. Magnifica Humanitas was developed over eighteen months, with input from scientists, ethicists, and, apparently, AI company co-founders who did not like what they read.
The encyclical’s final position is clear, if carefully worded. Machines cannot be conscious. Human dignity cannot be computed. And the industry’s self-serving argument that its products might be moral patients deserving of rights is exactly the kind of ideology the Church believes it exists to resist.
Whether that verdict holds is another matter. The question of machine consciousness is unlikely to be settled by any document, papal or otherwise. What Magnifica Humanitas and the confrontation surrounding it have done is drag a debate that was largely confined to AI safety conferences and philosophy departments into the center of global moral discourse.
That is, perhaps, exactly where it belongs.
