The Pope, a Chatbot, and the Chinese State

The Vatican has issued warnings about artificial intelligence. Anthropic has given its AI assistant a constitution. China has developed governance frameworks for intelligent systems.

This sounds less like a policy debate than the setup to a philosophical riddle. But beneath the differences in theology, Silicon Valley engineering, and state governance lies the same question: How do you shape an intelligence powerful enough to influence human thought, behaviour, and society?

For centuries, humans developed frameworks for governing intelligence. We wrote constitutions for nations, ethical codes for professions, curricula for schools, and moral traditions for civilizations. We assumed intelligence required formation, boundaries, and purposes.

Now we are doing the same for machines. That marks a profound shift in the technological story.

Most technologies amplified human muscle. The steam engine multiplied physical power. The assembly line accelerated production. Even computers, for much of their history, were understood primarily as tools for calculation, storage, and automation.

Artificial intelligence belongs to a different category altogether. It deals in language, judgement, persuasion, interpretation, and decision-making. It summarizes information, recommends actions, generates arguments, mimics empathy, and increasingly occupies territory once associated with teachers, advisers, experts, and institutions.

Steam engines did not require constitutions. Nobody worried that they might advise teenagers, influence elections, write textbooks, or shape a society's understanding of truth.

Consider the Vatican's concern. In his first encyclical, Pope Leo XIV enters the AI debate through the lens of human dignity. What happens when intelligence becomes detached from conscience? What happens when systems capable of extraordinary analysis remain incapable of moral responsibility? What happens when persons are increasingly reduced to profiles, predictions, and data points?

The Church approaches AI through one of humanity's oldest insights: intelligence without ethical orientation can become dangerous. Capability alone is not wisdom. Efficiency alone is not flourishing.

In January 2026, Anthropic, currently the most valuable AI company, published a constitution for Claude. Faced with the problem of aligning increasingly capable systems, engineers did not rely only on optimisation metrics or technical safeguards. They produced something resembling a moral framework. The assistant receives principles. It is expected to reason within them.

This is, in a curious way, an educational move.

Every constitution contains a hidden theory of judgement. It expresses assumptions about what kinds of reasoning should be cultivated, what trade-offs matter, what behaviours deserve encouragement or restraint.

A constitutional AI is not simply a safer chatbot. It is an attempt to teach a synthetic intelligence how to behave under uncertainty.

Then there is China.

China's governance approach emerges from a different institutional vocabulary, one focused more explicitly on stability, security, coordination, and sovereign oversight. Powerful intelligence, in this framework, must remain governable.

This is often presented in Western commentary as uniquely restrictive. But the contrast is not as simple as freedom versus control. Every society places constraints around intelligence. Every educational system rewards some forms of reasoning and discourages others. Every platform moderates some behaviours and amplifies others. Every institution carries assumptions about acceptable risk, acceptable authority, and acceptable autonomy.

China simply articulates questions of governability more directly than many liberal systems, which often distribute similar questions across markets, corporations, and digital platforms.

Seen together, the Pope, Anthropic, and China are not simply offering different answers. They are measuring AI through different moral currencies.

The Vatican foregrounds dignity, meaning, and moral responsibility. Anthropic foregrounds safety, behavioural alignment, and procedural principles. China foregrounds stability, coordination, and collective governability.

None of these frameworks is neutral. Each decides what matters most, and what risks are easier to overlook. Some harms fit neatly into dashboards, compliance systems, and quarterly reports. Others do not. Cognitive erosion does not appear in GDP statistics. The loss of meaning does not produce a clean performance metric. Human dignity cannot be plotted easily on a graph.

The disagreement over AI governance is therefore not merely about rules. It is about vision.

Each framework sees some dangers sharply and others dimly. The Pope notices the erosion of dignity and moral agency. The chatbot constitution notices alignment failures and behavioural harms. State governance notices instability and loss of control.

The challenge may not be to discover a single master metric for governing intelligence. Any framework that evaluates AI in a single currency is already making a philosophical choice about what counts and what can safely disappear from the ledger.

The harder task is wider vision. Economic efficiency sees some truths. Human dignity sees others. Safety engineering sees others still. No single lens captures the whole landscape.

Schools have wrestled with this problem for generations. Every educational system carries its own moral currency. Some optimise for employability. Some for citizenship. Some for examination performance. Some for creativity, discipline, or national cohesion. What goes unmeasured often slips toward invisibility.

AI governance is beginning to confront a similar dilemma.

Because teaching intelligence always requires a prior decision about what intelligence is for. Productivity? Truth-seeking? Autonomy? Social stability? Human flourishing?

We often speak about "aligning AI," as though alignment were principally a technical problem. But alignment presumes a destination.

The deeper difficulty may be that societies possess far less agreement than we imagine about what good intelligence looks like. The problem may not be that machines lack a settled purpose. It is that humans do.