Can AI really become controllable? Aleteia questions the myth of control

Aleteia publishes a reflection on the technical and anthropological limits of the project to make artificial intelligence controllable. The magisterium of Leo XIV already sheds light on the question.

The fact

On July 24, 2026, Aleteia FR published an article questioning the recurring promise of artificial intelligence laboratories to eventually offer "controllable" systems. The question is not new: the Nobel laureates gathered in Castel Gandolfo in July 2026 had identified this point in the Rome Declaration (publication #1177). The Aleteia article refocuses the debate on the very nature of technical control, highlighting the gap between the claimed algorithmic mastery and the reality of systems that self-optimize outside any humanly readable framework.

Our reading

The technical debate cannot be separated from its anthropological dimension. Leon XIV, in the encyclical Magnifica humanitas, set the framework: the properly human decision belongs to the nature of the person and is not delegated to a machine. Ballarani, from the University of Pavia, interviewed during the Rome Declaration, had summarized this line: "Man must never cede the power to decide to machines." This is not a principle of technical precaution but a principle of Thomistic realism. Aristotle had already taught, in the Nicomachean Ethics Book III, that the human act is what proceeds from an internal principle knowing the singular circumstances. A machine does not deliberate in the strong sense: it calculates. The question of "control" is therefore poorly posed when it is reduced to an engineering problem. It is a question of subsidiarity: what man can decide, man must decide, precisely because he is irreplaceable.

To ponder

The digital-sovereignty-europe thread is enriched with a new milestone. To the technicians, the responsibility of making their systems readable; to the faithful, the responsibility of refusing the abdication of judgment.