The Vatican’s first encyclical on AI asks who should govern the technology in the common interest. Reading it as an argument about institutions rather than theology, Vsevolod Shabad argues that every secular candidate – above all the state – is structurally conflicted, and that the impartiality AI governance needs is not a property to be found but an institution to be built.
On 25 May 2026 Pope Leo XIV published Magnifica Humanitas, the first papal encyclical devoted to artificial intelligence. He signed it on 15 May, the 135th anniversary of Rerum Novarum, Leo XIII’s 1891 letter on labour and capital, and presented it alongside a co-founder of one of the companies building it. The symbolism was hard to miss: an institution that speaks in centuries, sharing a stage with an industry that deals in weeks.
Strip away the theology and the document makes a sharp secular claim. It argues that governing AI in the common interest requires an impartial authority, and doubts that the obvious candidates can supply one. That is not a religious worry but a problem in institutional design.
The conflicted referee
The encyclical rules out the market acting alone. “It is no longer possible to rely solely on the invisible hand of the market,” it states; “politics has the task of orientating economies and technologies to the common good.” And the market’s stake is deepening, not dissolving. AI companies are moving into public ownership, acquiring a duty to maximise returns for shareholders. A player bound to win cannot also referee. The natural inference is that the state should referee instead. But it is not a neutral party either.
A government holds multiple roles at once. It is the referee, writing the rules. It is the beneficiary, taxing the profits and productivity that AI generates, so that slowing the technology is fiscally painful. It is the customer, procuring AI for public services and defence. And it is the payer of what the technology pushes off the private ledger: unemployment support, retraining, the health service that picks up the consequences of a lost livelihood. Those bills arrive at the Treasury, not at the company. That last role is not hypothetical. I have argued in this Review that when firms announce AI-driven redundancies, they capture the efficiency gains for shareholders and transfer the social costs to the public balance sheet.
A referee who also collects the gate receipts, buys tickets and pays the losing team’s wages is not a referee. Each role is legitimate on its own. Together they make impartial judgment structurally impossible.
The short clock
The second problem is about time. The alternation of power is one of democracy’s great virtues. But no arrangement delivers a good thing for nothing. The price of alternation is that nothing settled in one parliament is guaranteed to survive the next. And a technology that reshapes a country over decades needs a position that holds.
The United States shows the cost most plainly. An executive order on AI safety and oversight, signed by President Joe Biden in 2023, was rescinded in 2025 at the start of Donald Trump’s second administration and replaced with one focused on removing barriers. The point is the reversal, not the direction: same country, opposite policy, one election apart.
Britain shows the same compression from another angle. It has had seven prime ministers since 2015; the most recent, Andy Burnham, took office in July without a general election. One of the seven lasted seven weeks: a 2022 budget of unfunded tax cuts sent borrowing costs up so fast that the Bank of England had to step in to stop pension funds from failing, and the policy, the chancellor and the prime minister were all gone within a month. In democracies, the governing clock runs far shorter than the horizon over which AI reshapes labour markets and security.
This is not an argument for fewer elections, but one that democracies have already accepted elsewhere. Certain functions are deliberately placed beyond the cycle’s easy reach: interest rates sit with an independent central bank, fundamental rights with the courts. That insulation is never absolute – governments appoint the judges, set the central bank’s mandate and can legislate over both. But that is the point, not the objection to it. Insulation is not a natural property of these institutions. It was built deliberately, against resistance, and it lasts only as long as it is defended.
Pope Leo’s encyclical names the same need. The state must still regulate AI; the encyclical insists on that. What the state cannot be, by itself, is the patient and disinterested arbiter the task requires.
How disinterestness is built
Here the encyclical’s choice of anniversary becomes an argument rather than a flourish. Why can an institution two millennia old speak about concentrated technological power with any claim to disinterest? The answer is not that the Church is virtuous. It is that it lost the interests that would have compromised it.
In 1870 the papacy lost the Papal States – the band of central Italy, Rome included, that popes had ruled for a thousand years – to the armies unifying the country. With the territory went the tax base, the courts, the army and the seat among the powers of Europe. Twenty-one years later the same office issued Rerum Novarum, taking labour’s side against unchecked capital.
Whether it gave that power up by defeat or by choice does not matter; the sequence does. The standing to speak arrived only once the competing stake was gone. Disinterestedness was not a starting virtue but a residue.
The encyclical even supplies the template. It points to bodies built to protect the vulnerable: the United Nations, founded in 1945 by states that had just finished fighting each other and wanted somewhere neutral to argue, and the Red Cross, set up in Geneva in 1863, whose “operational neutrality” lets it care for all sides. Their standing comes from having no stake in the quarrels they enter.
The Red Cross was designed this way, and it shapes the fighting that it cannot command: within a year it persuaded governments to sign the first Geneva Convention, and the modern conventions name it as the body that visits prisoners of war. So impartiality can be engineered from the start, not only lost into existence as it was for the Church.
Building the referee
The provocation of Magnifica Humanitas, then, is institutional, not theological: the market is a player, and the state is referee, beneficiary, customer and payer at once, on a clock too short for the task.
The conclusion is not that an unelected, centuries-old institution should govern AI. The encyclical forecloses that reading: the Church “does not claim to assume the functions belonging to the State.” It is that democratic societies will have to build what the Church acquired by historical accident and the Red Cross acquired by design: bodies that sit outside the election cycle, take no side in how fast AI advances and can hold a line for decades.
A fair objection follows: a body that cannot tax, procure or punish has no power at all. But coercion was never the point. A referee makes the call; it does not drag players off the pitch. The instrument of a disinterested body is the nudge: the standards, benchmarks, disclosure rules and defaults that shape how states and firms build AI.
Its authority would be informational, not coercive. That is how technical and accounting standards work: an independent body writes the benchmark, and the state – through law, procurement and liability – and the market make it bite. The separation is the whole mechanism. Whoever writes the measure must not be whoever is judged by it, and preferably not whoever enforces it either.
The bodies we have so far – the European Union’s AI Office and the various national AI security institutes – come close. Compliance with Europe’s transparency rules for AI systems runs through a voluntary code on the marking of AI-generated content, drafted with providers, civil-society organisations and academics, and signed by some 190 organisations.
But that code was convened by the European Commission’s own AI Office. The Commission and its AI Board then certified it. And the AI Office now enforces the Act “together with national authorities”. Convenor, certifier and enforcer are a single institution – funded and staffed by the same states that tax, buy and underwrite the technology.
This is where disinterest does its work. It is what separates a legitimate nudge from manipulation: a steer from a conflicted party is a thumb on the scale; from a body with no stake, it is choice architecture. Building that architecture is work the engineering, standards and governance professions can begin now, without waiting for a treaty. The code proves that they will do it. What is missing is the wall between the people who write the measure and the people who wield it.
This article gives the views of the author, not the position of LSE Business Review or the London School of Economics. You are agreeing with our comment policy when you leave a comment.
Image credit: Butusova Elena provided by Shutterstock.
Facts Only
* Pope Leo XIV published Magnifica Humanitas on May 25, 2026.
* The encyclical addresses who should govern AI in the common interest.
* The document rules out relying solely on the invisible hand of the market for governance.
* Politics has the task of orienting economies and technologies to the common good.
* A government holds multiple roles: referee, beneficiary, customer, and payer.
* The author argues that combining these roles makes impartial judgment structurally impossible.
* The alternation of power in democracies is too short for long-term technological shifts.
* An executive order on AI safety was rescinded in the US within two elections (2023 to 2025).
* A UK government experienced rapid policy shifts during a short period involving changes in leadership and economic mandates.
* The text references the United Nations and the Red Cross as examples of bodies designed for operational neutrality.
* The European Union's AI Office enforces transparency rules alongside national authorities.
Executive Summary
The first papal encyclical on artificial intelligence suggests that governing the technology in the common interest requires an impartial authority, as neither the market nor obvious political candidates can provide it alone. The document frames the issue not as theology but as a problem of institutional design. It argues that the state, while possessing multiple conflicting roles—as referee, beneficiary, customer, and payer—is structurally unable to be a disinterested arbiter due to these overlapping responsibilities.
The text further details the temporal challenge: the rapid pace of technological change conflicts with the slow cycle of democratic alternation of power. Examples from the United States and the United Kingdom illustrate how policy reversals occur rapidly across elections, suggesting that insulating certain functions from the political cycle is necessary for long-term stability. The solution proposed is not an unelected governing body but the deliberate building of institutions—like those inspired by the Red Cross or the United Nations—that are designed to possess operational neutrality, separating their authority from immediate electoral pressures.
Finally, impartiality is achieved through informational authority rather than coercive power. A disinterested body would establish standards, benchmarks, and defaults that shape AI development, allowing states and markets to implement these rules, thereby creating a mechanism where separation between the rule-makers and enforcers ensures fairness, as demonstrated by existing voluntary codes on AI content marking.
Full Take
The central argument pivots on the impossibility of finding a disinterested authority—whether religious or secular—capable of governing rapidly evolving technology, given the conflicted roles held by the state and the market. The piece suggests that institutional impartiality is not an inherent property but a constructed artifact built out of historical shifts, analogous to how the Church lost temporal power by defeat, or how neutral bodies like the Red Cross were designed from inception to maintain neutrality.
The pattern observed here involves resisting the assumption that existing institutions can simply absorb new functions (like AI governance). Instead, it calls for engineering external, insulated structures—entities whose authority is informational rather than coercive—to manage the systemic risks created by technological acceleration. The shift from seeking inherent moral virtue (the Church's historical claim) to designing functional separation demonstrates a profound pattern of institutional response to power concentration.
The implications point toward cognitive sovereignty: true governance requires creating structural distance between those who set standards and those who execute them, preventing capture by the very systems they are meant to regulate. The observation that compliance codes can be managed by an institution that is itself funded by the conflicting actors suggests a critique of centralized enforcement, reinforcing the need for distributed, deliberately neutral mechanisms when dealing with systems operating on a timescale faster than political cycles.