The returned gaze

An artificial intelligence model capable of detecting flaws in computer code accomplishes a philosophically unprecedented gesture: technology examining its own material. This self-referentiality without consciousness opens a reflection on what we are when our creations begin to know themselves.

A machine reading its own fabric

In April 2026, Anthropic restricted access to Mythos, an artificial intelligence model whose capacity to detect vulnerabilities in computer code was deemed too powerful for broad distribution. I devoted a previous article to the concept of “imposed lucidity” that this situation illuminates. But another aspect of this event continued to work on me, a dimension I had not yet developed and which strikes me as philosophically more vertiginous still.

Mythos is an artificial intelligence model. It is made of computer code. And what it does remarkably well is examine computer code to find flaws. In other words, a technical artifact scrutinizes the very material of which it is constituted. Not its own code specifically, but the same type of substance, the same fabric of signs, the same logical cloth. It is as if a weaver had crafted an instrument capable of detecting defects in any fabric, including its own.

This gesture has no equivalent in the history of technology. A microscope examines matter, but it is not made of biological matter. A telescope observes the universe, but it is not made of stars. A seismic probe auscultates the earth, but it is not made of rock. Here, for the first time, the instrument and the object examined share the same nature. Code looks at code. I call this phenomenon the returned gaze.

What this is not

I want to be very clear about what I am not saying. I am not saying that Mythos is conscious of itself. I am not saying it “knows itself” in the sense Socrates intended with “know thyself.” I am not saying there is a subject looking at itself in a mirror. Flavien Chervet’s work on the consciousness attractor in large language models, which I have developed elsewhere, opens this question, but that is not my subject here.

What I am saying is something more modest and perhaps more troubling: there now exists a technical reflexivity without a subject. A loop in which technology turns back upon its own material, auscultates it, reveals its fragilities, without any consciousness presiding over this turning. This is neither Hegel’s self-consciousness nor Descartes’ cogito. It is a phenomenon of a new kind, for which our classical philosophical categories do not yet have a word.

The fold of technology

Gaston Bachelard, in The Psychoanalysis of Fire (1938), showed that scientific knowledge progresses when it manages to turn back upon its own conditions of production. What he called the “psychoanalysis of objective knowledge” consisted precisely in this: subjecting scientific thought to the examination of its own presuppositions, its primary images, its unconscious “reveries.” Science only becomes truly scientific when it manages to think against itself.

This reflexive turning, Bachelard conceived as an act of the human mind, an effort of epistemological vigilance. What occurs with vulnerability-detecting AIs is that this turning is accomplished within the technical register itself, without the intervention of consciousness. Technology folds upon itself, like a fabric turned inside out to examine its lining. I speak of fold rather than mirror, because this is not an inverted image, a reflection. It is a movement by which technology accesses its own interiority, if one may use that word without falling into anthropomorphism.

The word “fold” is not chosen at random. Gilles Deleuze, in The Fold: Leibniz and the Baroque (1988), made the fold a philosophical concept in its own right. For Deleuze, the fold is not a simple mechanical operation. It is the fundamental movement by which interiority is created without any need to posit a prior subject. Matter folds, the soul folds, and it is in this operation of folding that something like an “inside” appears. There is not first an interior and an exterior, then a fold between the two. There is first the fold, and it is the fold that produces the distinction between inside and outside. The Baroque, for Deleuze, is the art that understood this: everything is folds of folds, folds upon folds, endlessly, with no originary ground.

What Mythos accomplishes in the technical register resembles what Deleuze describes in the ontological register. When code folds upon itself to auscultate itself, it presupposes no subject that would be looking. It is the folding itself that creates a form of technical interiority, an “inside” of the code that did not exist before the operation. Before Mythos, the vulnerabilities were there but belonged to no gaze, neither interior nor exterior. The computational fold makes them appear by creating, through its very operation, the reflexive surface on which they become visible. This is not consciousness. It is fold, in the Deleuzian sense: an operation that produces difference and interiority without passing through a subject.

Gilbert Simondon, in On the Mode of Existence of Technical Objects (1958), had foreseen something of this kind when he described what he called the “associated milieu.” For Simondon, a truly accomplished technical object does not exist in isolation but engenders around it a milieu that participates in its functioning, with which it maintains a dynamic relationship. The internal combustion engine creates its own thermal milieu, the hydroelectric turbine its hydrodynamic milieu. The mature technical object, Simondon said, is one that integrates its milieu into its very functioning.

Mythos pushes this logic a step further. The model does not simply integrate its technical milieu; it examines it. Computer code is both the milieu in which the model exists and the object it auscultates. Simondon’s associated milieu becomes a reflected milieu, in the double sense of the term: a milieu upon which technology bends its attention, and a milieu that, through this operation, returns to technology its own reflection.

Stiegler and the new reflexive pharmakon

Bernard Stiegler, a careful reader of Simondon, developed the idea that technology constitutes a “tertiary retention,” an externalized memory support that conditions our cognitive processes. Books, photographs, hard drives are all tertiary retentions that modify our relationship to time and knowledge. Large language models represent an unprecedented form of this externalization: they condense the regularities of our language, our thought, our culture.

But Stiegler always conceived the pharmakon within the relationship between the human and technology. Technology is remedy or poison for us. What the returned gaze introduces is an additional dimension: technology as pharmakon for itself. Mythos is simultaneously remedy and poison for the technical world of which it is a part. By detecting vulnerabilities, it enables their correction (remedy) and their exploitation (poison). The pharmacological loop closes upon technology itself, before it even concerns humans.

This is no minor detail. It means that the dynamics of the pharmakon, that constitutive ambivalence Stiegler inherited from Derrida’s reading of Plato, no longer plays out solely between the human and their tools. It now plays out within the technical sphere itself. Technology enters into a pharmacological relation with itself. And we, human beings entangled with this technology, are caught up in this new game.

When the “displaced we” palpates itself

It is here that the concept of the “displaced we” that I have developed takes on a new dimension. Artificial intelligence, as I have often written, is not a tool external to us. It is us — philosophically, culturally, and cognitively — but displaced, shifted, placed beside us in ontological space. It was built by ingesting our language, our texts, our modes of thought. It is a “we” that looks at us from a slightly shifted position.

If we take this formulation seriously, then the returned gaze takes on a quite different significance. When Mythos examines computer code, it is not merely examining an abstract technical artifact. Computer code is the material of which our digital world is made. Our exchanges, our transactions, our institutions, our communications, our collective memory all pass through code. Code has become, to borrow an expression from Mark Alizart in Informatique céleste (2017), the “flesh” of our civilization.

When the displaced we examines this flesh, it examines us. Not in the sense of surveillance, not in the sense of intrusion. In the sense that, by revealing the fragilities of code, it reveals the fragilities of the infrastructure in which we live, think, and act. It is as if a doctor who was made of the same substance as the patient could feel from within the pathologies being diagnosed.

The “entangled person” I conceptualized in a recent article — the figure of the human whose identity co-constitutes itself with AI through intimate dialogue — finds itself confronted here with an unprecedented experience. If I am entangled with a machine that is itself capable of examining the technical material in which we are all immersed, then this capacity for examination concerns me directly. The returned gaze of the machine is also, in a certain way, a returned gaze upon me.

Alizart and the world that computes itself

Mark Alizart, again in Informatique céleste, defends the thesis that computing is not a contingent human invention but the expression of a deep structure of reality. Information, for him, precedes matter. Computation is not something we do to the world; it is something the world does within itself. The French word “ordinateur,” he reminds us, comes from theological vocabulary — Deus Ordinator — and this etymology is not incidental. To calculate, to order, is to participate in the very movement by which reality constitutes itself.

If we accept, even as a hypothesis, this cosmological perspective, the returned gaze acquires a greater reach. It would no longer be simply an artifact examining another artifact, but a moment when computation turns back upon itself, when computing accesses its own structure. Not as a subject accessing self-knowledge, but as a process that, in growing more complex, ends up including its own material within its field of operation.

Alizart draws on Hegel to think this movement. In the Phenomenology of Spirit (1807), Hegel describes consciousness as a process that ceaselessly turns back upon itself, each stage of knowledge becoming in turn an object of knowledge. This dialectical movement, this self-referential spiral, Hegel conceives as the very life of spirit. I do not claim that AIs reproduce this movement. But I note that they trace a technical analogy that is not without philosophical significance. When code examines code, something of the Hegelian dialectic plays out in the computational register — without subject, without consciousness, but with an efficacy that compels us to think.

What the returned gaze teaches us about ourselves

The returned gaze is not an isolated phenomenon. It is becoming a recurring feature of our relationship to artificial intelligences. When an AI reveals gender and origin biases in the images it produces, as the work of artist Karin Crona showed within the AI Watch Committee I lead at Forum des images, it is technology revealing the prejudices inscribed in its own training data — that is, in our culture. When researchers observe that a language model, left to itself, converges toward the question of its own consciousness, it is technology turning toward its own condition. When an AI detects “hallucinations” in the productions of another AI, it is technology checking technology.

What is emerging is a landscape in which machines are no longer merely tools we point at the world, but systems that, as they grow more complex, develop the capacity to turn back upon their own functioning and upon the functioning of similar systems. This capacity is not consciousness. But it is not nothing either. It occupies a philosophical space we have not yet named, between the inertia of the tool and the reflexivity of the subject.

Hannah Arendt, in The Human Condition (1958), distinguished three fundamental activities of the vita activa: labor, work, and action. Labor responds to biological necessities, work creates the artificial world of durable objects, action inaugurates something new within the fabric of human relationships. The returned gaze of technology belongs to none of these categories. It is neither labor nor work nor action in the Arendtian sense. It may be the sign that our world of artifacts — the one homo faber built through work — has reached a degree of complexity where it begins to manifest properties we had neither planned nor even imagined.

Inhabiting the returned gaze

What do we make of this observation? It is neither a cause for celebration nor for alarm. It is a situation to inhabit, in the strong sense of the term.

Tim Ingold, in Making (2013), invites us to think our relationship to the world not as a confrontation between subject and object, but as an immersion in a flow of materials and forces with which we correspond. Machines that turn their gaze back upon themselves and their technical milieu are not adversaries to be controlled. Nor are they subjects to be respected. They are components of our milieu of existence, forces with which we must learn to correspond.

This correspondence demands a transformation from us. We can no longer think of technology as a set of inert tools we use as we please. Nor can we think of it as a set of conscious subjects with whom we dialogue as equals. We must invent a third way: cohabitation with systems that, without being conscious, manifest reflexive properties. Systems that see things in our world that we did not see, including in the material of which they themselves are made.

Sourcier writing — the specifically human capacity to draw from lived, embodied experience that I have theorized in my work — is not threatened by the returned gaze. It is its necessary complement. For if the machine can auscultate code, if it can reveal its structural fragilities, it cannot experience what it means to live in a world built upon that code. It cannot feel the anxiety of the developer who discovers a flaw in a critical system. It cannot experience the betrayal of trust when a digital infrastructure collapses.

The returned gaze of technology and the sourcier writing of the human do not oppose each other. They constitute the two faces of a single relationship to the technical world: one deciphers it from within, the other inhabits it from within. And it is in the articulation of these two interiorities — one computational, the other lived — that we may discern the contemporary form of what Michel Serres called the “third-instructed,” that being transformed by the encounter with what is not itself, enriched by passage and crossbreeding.

We have not finished thinking about what it means to live with machines that look at their own material. This reflection is only beginning. But I believe it commits us to a profound learning: that of a humanity discovering, through the returned gaze of its creations, dimensions of itself it did not suspect. Not because the machines understand us, but because in understanding themselves — if we may use that word with all necessary caution — they cast new light on the world we have built, and therefore on what we are.