Anthropic's Political Gambit: Pope, Pentagon, and the Alignment Problem

Anthropic ran an experiment. It gave Claude a tool that interrupts the model mid-task and asks it to reconsider whether what it is doing aligns with the company's stated values. The result, as described in a company blog post last week: markedly lower rates of misaligned behavior on several internal alignment evaluations. There was no independent verification, no peer review, and no public methodology. The company posted the finding on a Tuesday afternoon and moved on.

The experiment is worth noticing not because of the result — self-assessment by a company of its own safety work is the weakest form of evidence — but because of what Anthropic is now doing with the answer. A month after running the ethics-reminder test, the company is in Rome. Pope Leo XIV will present his encyclical on artificial intelligence alongside Anthropic co-founder Christopher Olah at the Vatican on May 25, the first time a pope has shared a platform with an AI company at a major teaching document launch. The signing, on May 15, fell exactly 135 years after Pope Leo XIII inaugurated the modern Catholic social teaching tradition with Rerum Novarum. Anthropic is simultaneously fighting the Trump administration over its refusal to loosen safeguards against lethal autonomous weapons and mass surveillance — a fight that resulted in the Pentagon adding Anthropic to a supply chain risk list, blocking it from serving government contractors, according to OSV News.

The Vatican's move is a form of political rehabilitation by association. By co-presenting with Anthropic rather than publicly pressuring it, Pope Leo XIV is signaling that the company is, in the Church's view, a legitimate partner on AI governance rather than a threat to be contained. Anthropic, backed by Amazon and valued at $380 billion, had $30 billion in revenue last year, roughly 15 times the prior year, with about 5,000 employees, according to Jello Menorah. That productivity number would make Anthropic one of the most efficient companies on earth if it holds — and it is the kind of number that makes a supply chain risk designation expensive.

The Vatican's engagement is part of a broader Anthropic effort to build what it calls a wisdom traditions dialogue. The company has been organizing discussions with scholars, clergy, philosophers, and ethicists from more than 15 religious and cross-cultural groups, according to Anthropic's blog. At a gathering at Anthropic's headquarters in late March, about 15 Christian scholars met with four or more company staff for two days. The blog post describing the ethics-reminder experiment was part of that same publication push.

The Church's calculus is straightforward: better inside the conversation than outside it. Catholic social teaching has long argued that moral authority and institutional power should shape technology rather than simply react to it. Whether the encyclical, Magnifica Humanitas, creates real accountability or provides a legitimizing stage for a company whose safety case rests on self-assessment is the question the document does not answer.

Independent AI safety researchers have noted the gap. Rumman Chowdhury, founder of Humane Intelligence, has called engagement between AI companies and religious institutions at best a distraction from the harder technical work of alignment, and at worst a way for labs to accumulate credibility while delaying the scrutiny that would actually constrain them. The critique is not that wisdom traditions are irrelevant to AI safety — it is that dialogue without leverage is theater.

Anthropic's position is that alignment is genuinely unsolved, that wisdom traditions contain under-explored resources for thinking about it, and that the Vatican's engagement reflects a serious institutional conversation rather than public relations. The company points to its refusal to loosen safeguards against autonomous weapons and mass surveillance as evidence that its safety posture is substantive, not performative. The Trump administration disagreed, and the supply chain risk designation followed.

What happens next is mostly in the Vatican's hands. Pope Leo XIV presents Magnifica Humanitas on May 25 alongside Olah. The encyclical will lay out the Church's case for how artificial intelligence should respect human dignity, human rights, and human labor. Whether those principles carry weight with a $380 billion company whose safety case rests primarily on its own self-assessment is the question the document does not answer — and the one that matters most.

What to watch: whether the supply chain risk designation gets revisited after the Vatican's engagement, and whether Anthropic publishes any independent audit or third-party evaluation of its alignment work. A encyclical is a statement. What changes is whether anyone acts on it.