A company announces an AI strategy. The strategy becomes a principles document. The principles document is circulated, acknowledged, filed. And the actual decisions — what gets automated, whose autonomy shrinks, which data crosses which boundary — happen in the implementation, far from whoever signed the framework. The same gap — between the document and the decision — exists at national and international level.

Worldwide spending on artificial intelligence (AI) will reach $2.5 trillion this year. Not revenue — capital deployment. Data centres, model training, inference infrastructure, autonomous systems launched into orbit. The world’s first trillionaire was minted last week on the back of AI-adjacent infrastructure. These numbers exist to establish a single point: no moral argument, no governance framework, no institution acting alone can match the velocity of what is being built. The asymmetry between deployment and deliberation has never been wider.

On 25 May 2026, the Catholic Church deployed its highest form of doctrinal authority on the subject of artificial intelligence. Pope Leo XIV’s encyclical Magnifica Humanitas — the first in history devoted to AI — draws a deliberate structural parallel to Rerum Novarum, the 1891 encyclical that responded to the Industrial Revolution and established Catholic social doctrine on labour and dignity.

The framing in the encyclical is anthropological, not technological. The document does not ask whether AI will harm us. It asks whether we will remain recognisable to ourselves. At the presentation launching the encyclical, a co-founder of Anthropic said publicly that frontier AI labs operate inside incentives that conflict with doing the right thing, and asked for critics from outside those incentive structures.

This is worth noting not because of who said it, but because of what it reveals about the structural position of the people building. They are inside a machine whose logic they recognise as insufficient. They are asking for external constraint — not from regulation alone, but from traditions of thought that predate the incentive structures they inhabit. I wrote last October that institutions like the Church offer something Silicon Valley lacks: the capacity to think in centuries rather than quarters.

That framing was incomplete.

The obvious objection: the Catholic Church suppressed science, centralised power, silenced dissent across centuries. Religious institutions that think in long timeframes also act in long timeframes — slow to acknowledge harm, slow to reform, capable of precisely the abuses of power they counsel others against. But the argument here is not about religious authority. It is about a specific mechanism.

A study published in Research Policy this year analysed 112 monasteries across Germany, Switzerland, and Austria — combining survey data, web presence analysis, and ethnographic observation in Benedictine communities. The finding contradicts the standard assumption that old organisations calcify. Monasteries with centuries-old co-determination structures — elected leadership, reversible authority, mandatory consultation on significant decisions — navigate digital transformation significantly better than those without.

The study examined digitalisation, not AI specifically. But the structural principle — that participatory adoption outperforms imposed adoption — is not contingent on the technology being simple.

The researchers use a term from evolutionary biology: exaptation. A trait that evolved for one purpose proving useful for an entirely different one. Dinosaur feathers evolved for temperature regulation and later enabled flight. Monastic co-determination evolved for spiritual communal governance and later enabled digital adaptation.

The mechanism is specific: participatory governance practised long enough to become cultural reflex rather than imposed policy. Not memory, not age, not tradition — decision architecture. Distributed authority, local responsibility, consultation as default.

New Zealand — my country — released a Public Service AI Framework this year. Academic researchers have characterised it as “Pollyanna policy”: non-binding principles, no enforcement mechanism, an assumption that existing institutional structures can absorb genuinely novel challenges without structural adaptation.

The framework declares values. It does not specify the mechanism by which those values survive the pressure to move fast.

This is not a problem unique to New Zealand. Most AI governance frameworks — corporate, national, international — follow the same pattern. They are principles documents. They describe desirable outcomes. They do not specify how those outcomes will be maintained when the pressure to move fast becomes overwhelming.

The monastery research illuminates the failure mode precisely. Principles without participatory mechanism produce governance theatre — the appearance of oversight without the structural capacity to exercise it. The framework exists. The decisions happen elsewhere.

The same dynamic plays out inside enterprises — and this is where the monastery finding becomes unexpectedly practical.

Every company deploying AI at scale faces a version of the governance question. The competitive pressure is real: move too slowly and your competitors gain structural cost advantages that compound weekly. Move too fast and you accumulate dependencies, privacy exposures, and productivity claims that nobody has validated. The temptation — overwhelmingly — is to resolve this tension through top-down deployment. A platform is chosen. A rollout is announced. Adoption is measured.

I wrote earlier this year about how individual leaders delegate confidence to vendors, models, process, enthusiasm. The same happens at organisational level — except the delegation is to a framework nobody follows. The monastery finding suggests a different model. Not slower — participatory. The people affected by the change have structural authority over how it gets adopted. Local, consultative, reversible. The question shifts from “how do we get people to use this?” to “what do the people who will live with this think it should do?”

The drivers differ — profit in one case, public service in another — but the structural failure is the same: decisions delegated to a framework rather than distributed to the people who bear the consequences. That is not a call to slow down. It is a structural observation about what kind of adoption produces durable change rather than shallow compliance.

A technology that reshapes labour, knowledge, creativity, warfare, and the structure of human relationships is being deployed at unprecedented speed. Models that achieve gold-medal standard at the International Mathematical Olympiad. Autonomous agents deployed in production. Infrastructure launched into orbit on reusable rockets.

Against this: governance frameworks that are largely voluntary, largely written by the same institutions that benefit from minimal constraint, and largely lacking the participatory mechanisms that — according to the only empirical research we have on institutions surviving centuries of technological disruption — actually produce adaptive capacity.

The mechanism requires what no single tradition provides: legislative power, empirical discipline, advocacy reach, moral imagination, and the creative capacity to say what it means to be human. Operating simultaneously. In tension. Without resolution.

The alternative is what we have now: principles documents that nobody is obliged to follow, and decisions that happen elsewhere.

The slow questions still need asking. But there is a harder admission underneath: most of us are not asking them. We are deploying. We are adopting. We are inside the machine whose logic we recognise as insufficient — and the asking, if it happens at all, we have delegated to someone else.

References