AI That Improves Itself: Anthropic Facing the Abyss It Helped Open

Verified Raw Facts

Anthropic has issued a warning about the risk of the emergence of an artificial intelligence capable of autonomous self-improvement, without direct human intervention. The company consequently advocates for global AI governance.

The conference held by the Institut Catholique de Paris (ICP) on June 11, 2026, brought together philosophers, jurists, and ethicists for a rereading of Magnifica Humanitas, the encyclical by Leo XIV. Among the speakers: Eric Charmetant (Loyola Paris Faculty of Philosophy), Diane Galbois-Lehalle (Chair of Digital and Citizenship, ICP), and Emmanuel R. Goffi (Applied Ethics).

A reflection published on the encounter between the Church and Silicon Valley recalls: "The word 'trasumanar,' which gave rise to transhumanism, comes from Dante's Divine Comedy, to express entry into eternal life. I do not believe that humanity's salvation lies in bioelectronic enhancement."

Analysis of Underlying Issues

"Recursive self-improvement" refers to a process by which an AI modifies its own parameters to enhance its future performance. Beyond a certain threshold, these modifications become unpredictable to human designers. The AI "learns" things its creators did not explicitly teach it.

Anthropic, which builds systems on the verge of this threshold, appears to believe that transparency is preferable to silence. This is noteworthy. Yet the structural irony remains: the warning comes from within the technological race itself. Everyone is braking while moving forward.

Magnifica Humanitas by Leo XIV (2026) had anticipated this: "Technology is not neutral: it carries within it the vision of the human being who designs it." An AI that improves itself is an AI that, at least partially, escapes this vision. In a sense, it becomes the author of its own purpose.

Philosophical Insight

Fabrice Hadjadj reminded us that the great temptation of our time is to believe that humans can surpass themselves through technology. Dante called this movement "trasumanar"—but in the Comedy, this transcendence is the fruit of a received gift, a grace, not a calculated performance.

In Le Règne de l'homme (Gallimard, 2015), Rémi Brague shows that the modern project of dominating nature carries a fundamental contradiction: by becoming master of everything, humans risk losing the very notion of what they are.

Laudato Si’ (Francis, 2015, no. 106) formulates the diagnosis: "The direction taken by techno-economic progress does not adequately resolve global emergencies." This is not a rejection of technology. It is a reminder that technology without a clear human purpose is not progress—it is drift.

Avenues for Reader Reflection

Global AI governance is a necessary institutional response. It is insufficient. The true limit of self-improving AI is not legal. It is philosophical: to recall that humans are not functions to be optimized, that the human person has a dignity that cannot be reduced to measurable capacities.

This is precisely what Magnifica Humanitas states. And this is what the ICP philosophers debated on June 11. The reception of the encyclical in academic circles is not trivial: it is a sign that Rome still speaks to the intellectual world. Provided that this dialogue continues and does not remain confined to conference rooms.