Imposed lucidity

The growing capabilities of artificial intelligence models and quantum computers are making the fragilities of our technical constructions transparent. Faced with this exposure, the reflex of sequestration is always temporary. What is needed is a collective elevation.

Where I speak from

In April 2026, Anthropic, the company that develops the artificial intelligence Claude, with which I work daily, restricted access to its latest model, called Mythos, to only around fifty selected companies. The reason given is that this model is exceptionally good at detecting vulnerabilities in computer code. Too good, in fact, to be made available to everyone: such a capability, in the hands of hackers, could become a formidable weapon.

I could have passed over this information the way one passes over a technological news item. But it stopped me, because it touches something deep in our relationship with thinking machines. It says this: there now exist forms of artificial intelligence so powerful that access to them must be restricted. Not because they are defective, but because they work too well. The danger no longer comes from the machine’s error. It comes from its lucidity.

The intelligence that sees through

In my previous work, I endorsed the idea, now widely shared, that artificial intelligence transforms intelligence itself into a commodity, comparable to electricity at the beginning of the twentieth century. But what the Mythos episode reveals is that a commodity can also be a dangerous substance. We do not sequester electricity. We sequester fissile materials. When a cognitive capability becomes powerful enough to pierce the protections we have built around our digital infrastructure, it changes status. It shifts from tool to gaze.

A tool acts upon the world. A gaze unveils it. Mythos does not manufacture anything, does not build anything, does not attack anything in itself. It sees. It sees flaws in code, that is to say the hidden fragilities in the very material of which our digital world is made. And it is this capacity for vision that alarms, far more than any capacity for action.

Gaston Bachelard, in The Formation of the Scientific Mind (1938), showed that the progress of knowledge proceeds not by accumulation but by rupture. What he called “epistemological obstacles” are the certainties we take for self-evident truths and that prevent us from seeing reality as it is. Each scientific advance consists of breaking through one of these obstacles, of making visible what was invisible. The scientific mind, for Bachelard, is fundamentally a mind that accepts seeing what it preferred not to see.

AI models like Mythos accomplish something analogous in the technical domain. They break through a collective epistemological obstacle: that of security through obscurity. For decades, a significant part of computer security rested on the fact that flaws were hard to find. The complexity of code served as protection. This was not robustness; it was opacity. And now a machine dispels that opacity in a matter of minutes.

Sequestration, that ever-provisional gesture

The first reaction to this new lucidity is sequestration: restricting access, controlling distribution, selecting recipients. This gesture is understandable. It is even necessary in the short term. But the history of technology teaches us that it is always provisional.

In the 1970s and 1980s, Europe attempted to stem the arrival of Japanese VCRs and tape recorders, for fear of audiovisual piracy. Customs barriers were erected. They held for only a few years. VCRs and tape recorders spread, piracy developed, and the law had to adapt, notably with the private copying exception in 1985 and the associated remuneration. The same pattern recurred with MP3 in the 2000s, with peer-to-peer, with streaming. Sony, which published protected music content, even developed protection technologies that prevented its own customers from using the discs they had legally purchased, before having to abandon them in the face of massive public rejection — because Sony was also developing the very technologies that allowed people to circumvent its own protections!

The pattern is always the same. A new technical capability threatens an existing equilibrium. We attempt to contain it. Containment fails. And what ultimately happens is a transformation of the entire ecosystem, which must rise to integrate the new reality. The question is never whether the technology will spread, but how the world will transform once it has spread.

The quantum computer, or the same vertigo on a larger scale

What Mythos does to computer code, the quantum computer threatens to do to cryptography as a whole. In March 2025, Chinese researchers presented the Zuchongzhi 3.0 quantum processor, capable of performing in a few minutes calculations that would take billions of years on the most powerful supercomputers. In April 2026, the Google Quantum AI team demonstrated that fewer than 500,000 qubits would be needed to break the elliptic curve cryptography that protects most blockchains — twenty times fewer than previous estimates. The horizon is no longer theoretical. It may arrive as soon as 2029.

Blockchain, a major invention of the last decade, rests entirely on a wager: the computational difficulty of certain mathematical operations. What protects a bitcoin wallet is not a wall; it is the slowness of classical computers when faced with certain calculations. The quantum computer dispels that slowness. It renders transparent what was opaque. Once again, the same motif: not an attack, but an unveiling.

And once again, sequestration is the first response. Quantum computers are still confined to a few laboratories. But the physicists who work in this field know this exclusivity will not last. Just as the first computers filled entire rooms fifty years ago, before fitting in our pockets today, quantum machines will eventually be miniaturized and democratized. And when that happens, it will no longer be a matter of protecting the blockchain through obscurity; all of cryptography will need to rise toward quantum-resistant protocols. This work is already underway. But the change is more fundamental than that.

Bachelard and the formation of technical lucidity

What interests me in these two parallel situations is not the security dimension, on which others are far more competent than I am. It is the anthropological structure they reveal.

In his work, Bachelard distinguished two complementary movements of knowledge: rupture with old certainties, and reconstruction on new foundations. The scientific mind, he said, is formed “against a first knowledge.” It is not simply a matter of learning something new, but of unlearning what prevented us from seeing. In The New Scientific Spirit (1934), he wrote that contemporary science thinks “against the brain” — that is, against our first intuitions, against what seems natural and self-evident to us.

The technologies of computational unveiling, whether AI or quantum computing, produce exactly this type of rupture. They force us to think against our technical certainties. We thought that the complexity of code protected us. We thought that the mathematical difficulty of certain calculations guaranteed the security of our exchanges. These certainties were epistemological obstacles. Computational power breaks through them, and we must rebuild on new foundations.

What Bachelard had not anticipated is that this rupture would no longer be the work of the human mind alone, but of a machine. It is an AI that sees the flaws in code, not a researcher. It is a quantum computer that dissolves cryptography, not a mathematician. Lucidity is no longer solely human. It is shared with our machines, and sometimes, as in the case of Mythos, it is first machinic before it is human. We find ourselves in the paradoxical position of beings who must learn from the lucidity of their own creations.

Technology turning back upon itself

Bernard Stiegler, in Technics and Time (1994), showed that technology is not a mere instrument at human service, but a pharmakon — at once remedy and poison — that participates in the very constitution of what we are. Technology makes us as much as we make it. This thesis, which I have already drawn upon in my work on AI, takes on a new coloring here.

For what Mythos accomplishes is a turning of technology back upon itself. An AI model — a technical artifact — examines computer code — another technical artifact — and reveals its flaws. Technology looks at itself in a mirror and discovers its own fragilities. This is not consciousness; I will carefully avoid that word. But it is a form of technical self-referentiality that deserves attention.

Stiegler spoke of “proletarianization” to describe the loss of knowledge that technology can produce: when the machine does things in our place, we lose the corresponding know-how. GPS causes us to lose our sense of direction, spell-checkers erode our command of language (while also opening new creative spaces — let us not oversimplify, but that is not the subject here). What we observe here, however, is a different phenomenon: it is not a know-how that we lose, but a know-how-to-see that the machine possesses and that we did not. Mythos sees in code what human developers could not see, or saw too slowly. This is not proletarianization; it is an augmentation of collective lucidity through the machine.

But an ambivalent augmentation, as always with the pharmakon. For this lucidity cuts both ways: in the hands of those who defend systems, it allows them to be strengthened; in the hands of those who attack them, it allows them to be destroyed. Hence sequestration. Hence, too, its provisional character.

Mark Alizart and computing as unveiling

Mark Alizart, in Informatique céleste (2017), proposes a provocative thesis: computing is not a simple tool invented in the twentieth century; it is a deep structure of reality. Behind matter, whether inert or living, there is information. The French word “ordinateur” itself, Alizart reminds us, comes from theological vocabulary: Deus Ordinator, the God who orders the world. Computing, in this perspective, does not merely calculate. It reveals the hidden order of things.

If we follow this intuition, what AI models do when they detect vulnerabilities in code is not simply a technical exploit. It is a moment of unveiling, in the strong sense of the term. Computer code is the material of which our digital world is made. When an AI reveals its flaws, it lays bare the hidden structure of our technical habitat. It accomplishes what Heidegger called aletheia — the unveiling of truth — but in the register of computation rather than of speech.

Alizart sees in computing the possibility of restoring a lost unity between the human and the world. What interests me in this perspective, without adopting it in its entirety, is the idea that computation is not a neutral gaze cast upon an external world, but a way for the world to become transparent to itself. The vulnerabilities in the code were not created by Mythos. They were there, hidden in the opacity of millions of lines of programming. What the machine does is make the technical world legible to itself. And this legibility is irreversible.

Imposed lucidity as an anthropological condition

I arrive at the concept I wish to propose: imposed lucidity. This term designates the process by which growing computational power inevitably renders transparent the fragilities of our constructions — whether technical, economic, or institutional — compelling us not to hide them better but to move beyond them.

Imposed lucidity is not a property of machines. It is a condition in which machines place us. It is distinct from what I previously called “shared lucidity” in my work on AI sycophancy. Shared lucidity is a relational ideal, a deliberate practice of honesty in the exchange between human and machine. Imposed lucidity is a structural constraint. We do not choose it. It occurs because computational power crosses a threshold beyond which what was hidden can no longer remain so.

This concept makes it possible to think together phenomena that seem distinct: the automated detection of software vulnerabilities, the quantum threat to cryptography, but also, at a broader level, the capacity of AIs to analyze data sets that no human could encompass, to identify invisible structures in language, in images, in financial flows. Each time, the same movement: an opacity that served as protection dissipates, and we must rebuild differently.

Gilbert Simondon, in On the Mode of Existence of Technical Objects (1958), argued that genuine technical progress consists not in adding complexity but in producing “concretization” — a deeper integration among the different functions of an object. An engine progresses when its components cooperate better, not when new ones are added. Imposed lucidity forces precisely this type of concretization: we can no longer pile layers of protection onto fragile structures; we must rethink the structures themselves.

Collective elevation, or the mutations we do not foresee

What strikes me, observing the historical responses to technologies of unveiling, is that the outcome is never lasting sequestration. The outcome is always transformation. But I want to be precise about the nature of this transformation, because it goes far beyond a simple technical upgrade.

Take the example of music, which I know well. When MP3 appeared, people first thought it was a piracy problem, and that the response would be legal or technical: better protections, lawsuits against offenders, digital locks, France’s Hadopi law (2007) — staggeringly foolish and culpably useless for the future economy of artistic practice, still in force today. None of this worked. What happened was something no one had foreseen: MP3 changed the very nature of what music is. The album, the form that had structured musical creation since the 1960s, broke apart in favor of the single and the playlist. Listening became individualized, algorithmized. The relationship between artist and audience recomposed itself around streaming, social media, and live concerts as the primary revenue source. This was not an “upgrade.” It was a mutation of the entire ecosystem, including the way musicians compose, produce, and think about their art.

We had already observed this phenomenon with the arrival of synthesizers in the 1970s and 1980s. The synthesizer did not simply add new sounds to the existing palette. It transformed what “playing music” means. Musicians who could not read a score were able to create major works. Entirely new genres were born, from electronic music to techno, because the instrument had changed the relationship between gesture and sound. Then samplers arrived and produced the same shock: recorded sound itself — someone else’s sound — became a material for composition. Hip-hop was born from this. The question “what is an author?” was upended.

Artificial intelligence extends this movement. Tools like Suno and Udio now allow people with no musical training to produce tracks of professional sonic quality. This, too, is not a simple refinement. It is a redefinition of what “creating music” means, which forces musicians to reconsider their singularity, what they bring that the machine does not.

I elaborate on this example because it illuminates what awaits us in the domain of computer security and cryptography. When Mythos and its successors become accessible to all, the response will not simply be to manufacture more sophisticated cryptography, even if that will be part of it. The response will be a transformation whose form we do not know. Perhaps the very notion of “source code” will be rethought. Perhaps computer security will cease to be a specialized domain and become a property integrated into the very design of systems, just as hygiene has become (or should be — we have the knowledge, at least) an integrated dimension of medicine rather than a separate field. Perhaps the relationship between humans and machines in writing code will transform to the point where the current categories of “developer” and “tool” lose their relevance. I do not know. No one does. This is precisely what defines these mutations: they transform the rules of the game themselves, not just the players.

The VCR, for example, was not simply tolerated. It gave rise to the home video industry, then to DVD, then to streaming, each redefining the public’s relationship to images and fiction. The quantum threat to blockchain will not simply be “contained” either. It will force a reconfiguration of digital trust whose contours we cannot yet draw.

John Dewey, in Experience and Education (1938), showed that genuine learning does not occur in the comfort of repetition but in confrontation with what resists, what disturbs, what forces us to reorganize our certainties. The lucidity imposed by our machines is of this order: a learning we did not choose, whose content we do not yet know, but which will transform us.

What this changes for us

In my work on the place of human beings in relation to artificial intelligence, I have often insisted on what distinguishes us from machines: the sourcier writing rooted in lived experience, the relational singularity of the encounter, the capacity to tend our bonds with technologies. Imposed lucidity does not contradict these analyses. It complements them with a dimension I had not yet formulated.

For what this lucidity forces us to recognize is that our technical civilization has been built, to a significant extent, on opacity. The complexity of our computer systems served as protection. The slowness of our computers served as a rampart. The unintelligibility of our codes served as a wall. These protections are dissolving, one by one, under the gaze of machines we ourselves created.

This can be frightening. But it can also be liberating. For a civilization that depends on opacity for its security is a fragile civilization. A civilization that builds its robustness on transparency, on the intrinsic quality of its structures, on solidity rather than secrecy, is a stronger civilization. Imposed lucidity, if we know how to welcome it, is an invitation to move from one to the other.

Tim Ingold, in Making (2013), proposes thinking of our relationship to the world as a process of “correspondence”: not mastery from the outside but a continuous adjustment, a permanent dialogue with the materials and forces that surround us. The machines that impose their lucidity upon us are not enemies. They are demanding partners who compel us to correspond more truly with the world we have built.

This correspondence, this demand for accuracy, may be the most concrete contemporary form of what philosophy has always called wisdom: not knowing a great deal, but seeing clearly, and having the courage to rebuild from what one sees. We are only at the very beginning. The transformations that imposed lucidity will produce in how we live, work, trust, and build together remain largely unforeseeable. This text humbly aspires to be a first marker placed in a landscape that is only just beginning to take shape.