The Vatican has entered the compute wars, and the strategy is fundamentally flawed.
With the release of the encyclical Magnifica Humanitas, Pope Leo XIV stepped into the Synod Hall alongside Anthropic co-founder Chris Olah to declare that artificial intelligence must be "disarmed." The 83-page document draws an explicit parallel between advanced neural networks and nuclear proliferation, demanding that algorithmic systems be stripped of their "armed competition" logic, removed from military-industrial supply chains, and heavily restricted by international regulatory cartels.
It is a beautifully written, deeply moral, and utterly naive manifesto.
By framing algorithmic development through the lens of kinetic weaponry, the Vatican falls victim to the lazy consensus sweeping global governance bodies. This consensus presumes that computation can be bottled, inspected, and pacified like enriched uranium. I have spent fifteen years building and deploying enterprise algorithmic frameworks across highly regulated sectors, watching institutions throw tens of millions of dollars at compliance theater. The hard truth nobody wants to admit is that treating software like a physical bomb does not secure human dignity; it merely consolidates power into the exact technocratic monopoly the Pope seeks to destroy.
The Flawed Physics of Software Disarmament
The fundamental error of the "disarm AI" position is a failure of technical definition. Nuclear disarmament works because weapons-grade plutonium requires massive, centralized, easily traceable industrial infrastructure. Centrifuges cannot be hidden in a basement or downloaded via an open-source torrent.
Machine learning is math, weights, and electricity.
To "disarm" an optimization function means halting the underlying mathematical iteration. You cannot lock a dual-use transformer model in a silo. A model trained to optimize logistics for a commercial delivery network can, with minor adjustments to its reward function, optimize flight paths for autonomous loitering munitions. The weights are just files—long strings of floating-point numbers.
When Magnifica Humanitas demands an "identifiable and verifiable chain of responsibility" that brings AI under public control, it ignores the structural reality of modern compute distribution. If a centralized authority attempts to enforce strict monitoring at the hardware layer—say, tracking every Nvidia H100 or B200 tensor core globally—they do not eliminate the competitive drive. They simply force the development underground, driving the most potent systems into sovereign black markets and unregulated jurisdictions where moral discernment is actively penalized.
The Centralization Paradox
The Pope’s encyclical heavily invokes the principle of subsidiarity, arguing that technological decisions should be made at the local level and that ordinary communities must not be reduced to passive recipients of corporate code.
Yet, the prescribed cure directly contradicts the diagnosis.
To enforce the sweeping, international regulations called for by the Vatican, you require an unprecedented, highly centralized global regulatory apparatus. You need an agency with the teeth to audit private codebases, seize servers, and dictate what architectures are permissible to build.
Look at what happens when you introduce massive regulatory burdens into software ecosystems. The European Union’s approach to technology regulation provides a clear case study: it did not democratize the market. Instead, it created an immense compliance moat that only trillion-dollar American conglomerates could afford to cross, effectively suffocating local digital infrastructure.
By demanding top-down, global oversight to ensure AI becomes a "common good," the Vatican is inadvertently volunteering to build the ultimate technocratic panopticon. True subsidiarity is not achieved by granting a centralized council the power to approve or deny software deployment. It is achieved through the messy, chaotic distribution of capability—specifically through decentralized, open-source architectures that allow local communities to run, modify, and own their weights locally without asking permission from a corporate board or a state bureaucrat.
The Perils of Ethics Wash Cartels
The optics of the Vatican press conference were highly revealing. Sitting next to the Pope was Chris Olah, whose company Anthropic pioneered the concept of "Constitutional AI"—training models to adhere to a specific set of written ethical principles.
While Anthropic’s technical research into mechanistic interpretability is world-class, the political alignment between ethical tech giants and global moral authorities should make everyone deeply uncomfortable.
When a handful of heavily funded labs partner with regulatory bodies to define what constitutes a "moral AI," they are not protecting humanity. They are forming an ethical cartel. They get to write the constitution that dictates the parameters of acceptable digital thought.
Magnifica Humanitas correctly notes that "a more moral AI is not enough if that morality is determined by a few." Yet, the document’s push for top-down international standards inevitably leads back to that exact room of elites. If a small council of corporate executives, theologians, and state officials lock down the frontier of research under the guise of safety, they don't eliminate bias. They simply institutionalize their own biases at the root level of the global digital stack, preventing anyone else from building alternative models that reflect different cultural, economic, or philosophical realities.
The Modern War Fallacy
The most urgent rhetoric in the encyclical targets the automation of conflict, flatly stating that it is impermissible to entrust lethal decisions to artificial systems. The text claims that the classical "just war" framework is completely outdated due to the automated scale of modern friction.
This is a profound misunderstanding of how algorithmic systems function in theater.
Imagine a scenario where a military commander must choose between two options to neutralize an adversary's command post in a densely populated urban area. Option A relies on legacy human analysis: a generic satellite map, manual target coordinates, and a massive unguided artillery barrage that carries a 30 percent margin of error and significant collateral risk. Option B utilizes an algorithmic sensor-fusion network that processes real-time telemetry, precisely mapping human shield positions and deploying a single, highly targeted strike with a sub-meter error margin.
Insisting on an ideological ban on automated decision-making tools in this context does not preserve human life; it increases the brutality of war. The machine does not replace human conscience; it provides the fidelity required for conscience to act accurately. By blanket-labeling the integration of algorithmic decision-making as an "abdication of responsibility," the Vatican creates an untenable moral stance that rewards tactical ignorance and guarantees higher rates of collateral destruction.
Distribute the Capability
The solution to the concentration of technological power is not the illusion of disarmament. The solution is radical proliferation.
If you are terrified of an opaque, corporate-controlled algorithm blocking access to healthcare, credit, or employment—a legitimate fear highlighted by the Pope—the answer is not to petition a government body to regulate that corporate algorithm. The corporation will simply hire an army of lobbyists, wrap their system in compliance paperwork, and continue operating exactly as before.
The only structural counterweight to a concentrated, weaponized technology is a distributed, accessible version of that exact same technology.
Instead of treating advanced computation like a nuclear weapon to be locked away, it must be treated like literacy. When printing presses spread across Europe, they disrupted the established hierarchy, triggered decades of intense social upheaval, and shattered the monopoly on information held by the institutional elites of the time. The solution was never to disarm the printing press to maintain a comfortable social order. The solution was to teach everyone how to read.
We are not dealing with a bomb. We are dealing with an accelerant for human thought. Trying to disarm it ensures that only the most ruthless, lawless actors will remain armed.
For a deeper look into the strategic realities of tech governance and the friction between open-source models and institutional control, listen to the comprehensive breakdown of the structural shifts under discussion: Tech Monopolies and the Regulatory Moat. This analysis highlights why centralized control structures struggle to adapt to decentralized software deployment.
