May 31, 2026

Magnifica Humanitas and the Question AI Cannot Answer for Itself

Assiduity AI

The deepest question about artificial intelligence is not whether systems can act intelligently.

It is whether their action remains answerable to a human purpose.

That question has been implicit throughout Losing the Thread. The series began by distinguishing objective pursuit from objective origin. A generative system may pursue a task, optimize a continuation, complete a workflow, or execute a sequence of actions. But it does not supply the purpose of the work. That purpose comes from outside the model: from users, institutions, policies, laws, moral judgment, and human communities.

Pope Leo XIV’s first encyclical, Magnifica Humanitas — “On Safeguarding the Human Person in the Time of Artificial Intelligence” — brings that distinction into the center of public debate. Released in May 2026, it was signed deliberately on the 135th anniversary of Leo XIII’s Rerum Novarum, the 1891 encyclical on labor and human dignity in the age of industrialization. The choice of date is part of the argument: Leo XIV reads artificial intelligence as this century’s upheaval, and he extends a long tradition of social teaching to meet it.

The encyclical is not a technical document, and it should not be read as one. Its force lies elsewhere. It asks what artificial intelligence is being made to serve. Its opening frames a choice between constructing a new Tower of Babel and building a city in which humanity can dwell together. That image organizes the moral landscape of the document. The encyclical insists that the human person cannot be reduced to productivity, data, optimization, or strategic advantage.

Notably, it does not treat technology as the enemy. Its premise is that technology is neither antagonistic to humanity nor inherently evil. But it is, in the encyclical’s framing, never neutral: it takes on the character of those who design, finance, regulate, and use it.

That premise is close to the argument of this series. The whole of Losing the Thread holds that a system’s behavior is shaped by an objective supplied from outside it. The encyclical makes the moral version of the same claim: a tool carries the purposes of the people who wield it. The technical question — whose objective governs the system, and does it survive execution? — is one operational face of the moral one.

That is not a separate debate from AI governance. It is the deeper version.

Where the series meets the encyclical — and where it does not

Much public discussion of generative AI still begins with capability. Can the system reason? Can it write, code, search, summarize, plan, and act? Those questions matter; capability determines what systems can attempt. But capability does not tell us what a system should serve. A more capable system can still be pointed at a bad objective, a poorly specified one, or an objective that gradually loses contact with the human purpose behind it.

This is why the reliability question cannot stop at performance. A model may perform well while the task drifts. It may remain fluent while weakening a threshold, softening an exception, broadening a mandate, or substituting a nearby objective for the one it was meant to preserve. In long outputs and agentic workflows, that drift compounds. The system can appear helpful while moving away from the purpose that justified its use.

That is a technical problem. It is also an institutional problem. At the limit, it is a human one.

Here, though, honesty requires marking the difference rather than papering it over. Magnifica Humanitas calls, in part, for restraint — for slowing the pace of development until human institutions can catch up. This series makes a different argument: that part of deploying AI responsibly is building the controls that let purpose survive the process. Those are not the same position. One urges deceleration; the other urges governable deployment. They can be held together — restraint and control are complementary, not contradictory — but they are distinct claims, and the series should not borrow the encyclical’s moral authority by pretending otherwise.

What the two share is a refusal to let “human-centered AI” remain a slogan. If AI systems are to serve human dignity, dignity cannot live only in a mission statement. It has to become operational: institutions need ways to specify what must be preserved, to monitor whether systems remain attached to it, and to produce evidence of how a system actually behaved. Otherwise, the phrase decorates systems whose real behavior is governed by local fluency, convenience, and scale.

This is the narrow place where the argument of Losing the Thread touches the broader moral debate — and it is worth being precise about how narrow it is.

The boundary of the claim

The series has not argued that runtime control is the answer to AI ethics. It is not. No control layer can determine the content of justice, dignity, prudence, or the common good. Those are human responsibilities. They require law, governance, moral judgment, institutional accountability, and public deliberation. A system that faithfully holds a given objective is no better than that objective; runtime control can keep a system attached to its purpose, but it cannot tell anyone whether the purpose is worth pursuing.

What runtime control can address is one necessary part of the problem: whether a system remains attached to the objective it has been given while it generates, reasons, summarizes, or acts.

That matters because purpose can be lost in motion. A policy can be clear at the start and weakened by the output. A human instruction can be precise in the prompt and diluted by local continuation. A rule can sit in the source material and still fade during synthesis. A workflow can begin with one mandate and end by completing a more convenient one. If we care what AI serves, the question is not only whether the right purpose was specified. It is whether that purpose governed the work throughout.

So the practical questions become institutional ones. Can an organization define the objective that should govern a system? Can the system remain in an active relationship to that objective while it works? Can the institution inspect whether the system held the thread?

A hospital may need a clinical constraint to remain binding. A regulator may need to apply a rule without quietly substituting administrative convenience. A school may need learning support without reducing students to predicted outputs. In each case, the human purpose has to govern the system rather than be lost in fluent execution.

These are technical architecture questions only on the surface. Underneath, they are about whether moral and institutional commitments can be held.

Holding the thread

This is where the language of holding the thread becomes more than a metaphor.

To hold the thread is to preserve the connection between capability and purpose.

It is to recognize that intelligence without orientation can still lose the way — that a system can grow more capable and, at the same time, drift further from what it was meant to serve. The problem of drift is not finally about better outputs. It is about preserving human authorship over the objectives that AI systems pursue.

The series closed on an economic argument: enterprise AI will not be judged only by how much it can produce, but by how much of that production can be trusted, reviewed, governed, and used. Magnifica Humanitas presses the same logic into a larger register. Technology will not be judged only by what it makes possible. It will be judged by what it makes of us — by whether the systems we build remain subordinate to human dignity rather than becoming instruments of domination, exclusion, or careless scale.

That does not require rejecting AI. It requires refusing to confuse capability with wisdom.

The serious path forward is neither panic nor blind acceleration. It is governance worthy of the power being deployed: the patient work of translating human purposes into institutional controls, legal obligations, technical constraints, audit records, and review practices. It is the discipline of asking — not once at the beginning, and not only after the fact — whether the system is still serving what it was meant to serve.

AI systems will continue to become more capable. They will write, reason, plan, and act across more of the world. That makes the question of purpose more important, not less.

The hardest question is not whether AI can pursue objectives.

It is whether we will build institutions capable of governing the objectives AI systems pursue.

This postscript follows Losing the Thread, a series on autoregressive drift, objective fidelity, runtime control, and the emerging governance layer in artificial intelligence.