Matthew Berman on Pope Leo XIV’s AI Encyclical and Anthropic’s Role

Pope Leo XIV has published Magnifica humanitas, an encyclical on safeguarding human dignity in the age of artificial intelligence. I want to examine some of its arguments and the debate around Anthropic’s participation in its presentation. The Vatican released the document on May 25, 2026; my discussion reflects that moment.

I read the title as a reminder that there is something special about human beings, even as people debate AI consciousness and the possibility that machines may replace human activities. The document warns that new technologies open possibilities whose long-term effects on dignity and the common good are difficult to predict.

The encyclical argues that technological power has become predominantly private, complicating public scrutiny and governance. I see that concentration in the small number of organizations shaping frontier AI. I also suspect Anthropic’s views influenced the document, although its representative’s presence at the launch does not by itself establish influence over its authorship.

I worry that regulation supported by incumbent AI companies can make it harder for startups to compete. I interpret Anthropic’s lobbying and restricted access to its cybersecurity model, which I call Mythos, as examples of a potential regulatory-capture problem. Those are my assessments of the company’s incentives, not proof of its motives or that regulation necessarily produces capture.

The Pope rejects a simple yes-or-no choice about technology and says its character reflects those who create, finance, regulate and use it. I appreciate that nuance: neither an outright stop nor unchecked acceleration addresses the question of who controls the technology and whose interests it serves.

The encyclical says human dignity does not depend on ability, wealth, status or a person’s choices. I worry that AI could disrupt the work and contributions through which many people form their identity, while also believing human qualities could become more important as AI spreads.

My own example is writing. I have adopted a policy against using AI to help generate original essays for our newsletter. Reading AI-generated prose made me value the raw ideas and even the imperfections of human writing more; that is my preference, not a general measurement of all AI-assisted work.

Leo argues that even AI designers have only limited understanding of how these systems function. He also maintains that AI may simulate language and empathy but lacks the affective, relational and spiritual perspective through which humans develop wisdom. This is the encyclical’s account of current AI, not a settled scientific test of machine consciousness.

I agree with the warning about AI companionship. The encyclical says simulated care can create the appearance of a relationship and might weaken the desire for human connection, especially where real bonds are already lacking. I am particularly concerned about young people forming attachments to characters, though AI conversation may also provide comfort to someone who is lonely.

I worry that if artificial companionship is easier than human relationships, some people will choose it instead. I connect this concern to declining birth rates, but that is my speculation about a possible social effect, not evidence that AI companionship causes the decline.

The Pope calls for AI to be “disarmed”: his document describes a competitive mentality extending beyond military uses into economic and geopolitical contests. I recognize that I also describe frontier labs and countries as being in an AI race. I welcome the moral challenge, while remaining skeptical that human competition will disappear.

In the encyclical, disarming does not mean rejecting technology; it means preventing domination, loosening monopolistic control and opening technology to public debate and diverse cultures. I think that is an important distinction. The document calls for appropriate regulatory tools, but also says regulation alone is insufficient.

I return to my criticism of Anthropic and worry that excessive regulation could entrench the few companies already leading the field. In my reading, there is a tension between regulation and broad participation if rules block competition. I see open source as one possible way to distribute control, but that is my interpretation rather than an explicit endorsement in the passages I discuss.

I find the encyclical more informed and nuanced than I expected. It cautions that claims about AI can quickly become outdated and challenges the assumption that technological progress automatically benefits everyone. I also highlight its discussion of attention-driven businesses that exploit addiction and insecurity.

Anthropic co-founder Chris Olah spoke at the Vatican presentation. I interpret his participation and the company’s approach to safety as self-serving positioning, and I speculate that the company seeks disproportionate influence. The Vatican’s announcement confirms that Olah was an invited speaker, but that alone does not establish a formal alliance or Anthropic’s influence over the encyclical.

In the excerpt, Olah says his interpretability team finds internal model structures resembling some neuroscience results, evidence of introspection and states that functionally resemble emotions. He explicitly says he does not know what those findings mean and calls for continued discernment. These research interpretations do not establish that an AI experiences emotion or consciousness.

I cite Brian Roemmele’s criticism that the company’s safety narrative could serve as a competitive barrier and a regulatory lever. I share that concern, but it is a criticism of incentives rather than a demonstrated account of why the Vatican invited Olah or what Anthropic intended. In his full remarks, Olah also called for informed outside critics of AI labs.

I then read David Sacks’s argument that broad government power over AI could enable censorship, surveillance and control, despite the shared goal of protecting human dignity. The concern illustrates another side of the governance debate: concentrating power in governments can raise questions just as concentrating it in companies can.

I discuss another commentator’s criticism of Anthropic’s public positioning. I do not agree with every example offered, including criticism of its advertising rivalry with OpenAI, but I agree with the concern that invoking model qualities and seeking policy influence may strengthen the company’s position. That remains a disputed interpretation, not proof of deception.

My overall reaction to the encyclical is positive, including parts with which I disagree: I appreciate its care and willingness to engage with difficult questions. My continuing objection is to what I see as Anthropic positioning itself as an authority over who should build AI. The question of how to distribute power over AI remains open.