Pope Leo XIV Reads AI as Babel; Rabbis Draw Three Lessons

An encyclical is not usually where I look for AI policy signals, but Pope Leo XIV's first one, Magnifica Humanitas, reaches straight for the Tower of Babel to describe what today's engineers are building. An essay in Tech Policy Press takes the metaphor seriously and reads it back through two thousand years of rabbinic commentary. The result is sharper than the usual beware-of-hubris homily, because the rabbis pull three distinct warnings out of the same story.

The first is about human objectification. A midrash from Pirkei d'Rabbi Eliezer imagines the tower grown so tall that a fallen brick made the builders weep while a fallen worker did not. The authors map that onto an AI economy already reshaping work for paralegals, coders, translators, illustrators and entry-level office workers, and onto harder edges like autonomous weapons and surveillance. They point to recent friction between Anthropic and the Pentagon over military uses as a preview of a permanent negotiation about when the apparatus is allowed to break the man for the sake of the brick.

The second warning is about concentration. In their reading, compute, capital and talent for advanced AI now sit with "two governments, a half-dozen companies and a handful of individuals," an arrangement they call unprecedented for a general-purpose technology. Nachmanides and Rabbi Jonathan Sacks appear as the counterweight, with Babel read as empire and plurality as the moral architecture of creation. That leads somewhere less expected. The authors argue against the dream of a single global AI framework, because the priorities of India, of Africa, of Europe's rights-driven regime and of a United States torn between accelerationists and local pushback are genuinely different, and a divergent regulatory map may be the last safeguard against any one actor writing the human future.

The third lesson is the subtlest. Rashi's midrash depicts confusion at Babel as one builder asking for an axe and being handed a shovel, with violence following. The authors extend that to systems we build but no longer follow, "optimized, efficient, but completely opaque." The classic AI-safety worry is misalignment; theirs is incomprehensibility, a world of markets and institutions designed by intelligences whose reasoning we cannot legibly share, and therefore cannot govern, correct or meaningfully consent to.

The honest caveat is that this is a moral essay, not a policy paper. It does not name the two governments or the half-dozen companies, does not describe how the Anthropic-Pentagon friction was resolved, and does not propose specific rules. What it does leave behind is a vocabulary of objectification, concentration and incomprehensibility that a regulator or a lab safety team could actually reach for when the next hard call arrives.