Can We Disarm AI?

WHEN POPE LEO XIV published “Magnifica Humanitas” in May, what struck me most was a verb. The Pope reaches twice for the same words—“to disarm”—as he reflects on artificial intelligence and the civilization we are building. As someone who spent more than a decade working on nuclear disarmament, including helping to build the campaign that produced the Treaty on the Prohibition of Nuclear Weapons, that verb landed differently for me than it might for most readers.

This spring, I signed the “Pro-Human AI Declaration,” a document that calls for centering human dignity and equity at the heart of AI governance. Having worked inside the nuclear disarmament framework, I want to offer an honest accounting: what it can teach us, where the analogy breaks down, and why people of faith may be essential to making any of this operational.

The nuclear governance regime rests on three pillars:

1. the Nuclear Non-Proliferation Treaty, negotiated in the late 1960s after the Cuban Missile Crisis made plain how close we had come to catastrophe; 
2. the International Atomic Energy Agency, which verifies compliance through international inspections; 
3. the Treaty on the Prohibition of Nuclear Weapons, adopted in 2017, which finally did what the NPT never managed—it banned nuclear weapons outright, modeling itself on other prohibitions of weapons of mass destruction and humanitarian disarmament treaties, such as the landmines and cluster munitions conventions.

The NPT of the 1960s was a “grand bargain,” where the five nuclear-armed states agreed to disarm in exchange for non-nuclear states promising not to develop weapons. But decades passed, and the nuclear states kept deferring reduction and elimination of their weapons. The result was a nuclear governance regime that had concrete verification machinery and solid diplomatic infrastructure but also normalized the very thing it claimed to eliminate. That’s why the 2017 treaty was needed as a complement.

Can this be translated to AI? The most useful lesson from the nuclear framework is the TPNW. When we launched the campaign that became the treaty banning nuclear weapons, I first thought it was a strange idea. What is the point of banning weapons if the countries that possess them refuse to join? But the logic was about changing behavior and norms, not just establishing hard law. International law works as a social process just as much as domestic law such as smoking bans.

The landmines treaty, even without U.S. or Russian signatures, changed how people viewed those weapons, reducing their use even in nonsignatory countries by making them morally expensive. The same dynamic has reshaped cluster munitions. Despite recent challenges to these treaties as some countries discuss developing landmines and cluster munitions again, the last few decades have seen significant drops in production, use, and victims of these indiscriminate weapons.

For AI governance, it matters enormously that we develop laws to progressively shape social norms. We cannot wait for the major tech companies to regulate themselves or the largest countries to lead the way. We need to start acting now. Coalitions of countries, civil society, and the private sector can move to regulate specific applications of AI that violate human dignity. As we make AI use and development socially, economically, and morally costly, we can shift the context in which powerful actors operate.

The IAEA model is also instructive: an independent international body empowered to inspect, verify, and set standards, connecting scientists and regulators across borders. An equivalent international agency for AI, to audit algorithmic systems, establish transparency requirements, and coordinate national regulatory authorities, for example, could provide the institutional backbone that current governance conversations lack.

The main challenge is that while nuclear weapons are made by nation-states, AI is made by private companies. The IAEA’s authority derives from nation-states granting it access, and even then, the nuclear-armed states have never allowed inspectors into their primary weapons facilities.

In addition, the NPT and the IAEA set up a two-tier system where one set of rules apply to non-nuclear armed states and another, less restrictive set applies to nuclear-armed states. For governing AI, we must not reproduce the same asymmetry of robust accountability for smaller actors, and none at all for the most powerful.

However, the private ownership nature of AI development could also be an opening. Companies need markets and operate in diverse legal systems across borders. Moves on AI regulation by the European Union have shown that even major U.S. and Chinese private firms might need to adjust their systems to remain inside a large market. Smaller countries that struggle to control tech giants can join coalitions that set conditions of market entry. That leverage is genuinely different from anything that existed in the nuclear weapons regulatory world.

We need to act before we lock ourselves into systems we cannot undo.

In “Magnifica Humanitas,” Pope Leo grounds AI regulation in the Catholic Church’s social doctrine: human dignity, the common good, subsidiarity, solidarity, and the universal destination of goods (meaning that private property must still serve a universal good). Applied to AI, that last principle means algorithms, data, and digital infrastructure cannot be treated as purely private property when they shape the conditions of everyone’s life. The AI systems transforming our world were trained on knowledge accumulated by all of humanity. That ought to mean something about who governs them and who benefits.

The conversations that produced the “Pro-Human AI Declaration,” released in March, were driven by similar convictions. The declaration is not merely a statement of concern; it is an assertion that this technology belongs in some meaningful sense to all of us, and that the communities least consulted in AI development are the ones most at risk from its misuse. Operationalizing the Pope’s encyclical means translating these moral commitments into specific institutional demands: transparency requirements, equity conditions, meaningful human oversight, and international standards that no company can simply opt out of by relocating servers.

What made the Treaty on the Prohibition of Nuclear Weapons possible was not a legal innovation. It was a shift in moral imagination, substantially driven by civil society and people of faith who insisted that weapons capable of annihilating cities were simply incompatible with human dignity, and who refused to accept that permanence as inevitable. Pope Francis was instrumental in this when he expanded the Catholic Church’s position on nuclear weapons to include a rejection of nuclear deterrence itself—not just the use of nuclear weapons.

Pope Leo is calling for something similar. He is asking us not to passively absorb the world that a handful of powerful actors are building, but to assert a different vision. People of faith bring something distinctive to this moment: communities, relationships of trust, and a vocabulary of the sacred that insists human beings cannot be reduced to data points or optimized away. We have stood against weapons of mass destruction by insisting on the infinite dignity of every person. That same conviction that human life has a value no algorithm can calculate is what AI governance most needs.

We need to act before we lock ourselves into systems we cannot undo. What I learned in the nuclear world is that the right moment for action is never perfectly convenient, and waiting only reduces your own power and agency.

Pope Leo has named the challenge. The “Pro-Human AI Declaration” has laid out the commitments. The nuclear framework, for all its limitations, shows us that international governance of technologies with catastrophic potential is possible when enough of the world decides it must be.