For centuries, producing false evidence required effort. It involved falsifying a document, imitating a signature, fabricating a photograph, convincing a witness, or corrupting a chain of custody. Forgery had obviously always existed, but it came at a price.

Artificial intelligence is radically altering this economy. It reduces the cost of producing believable things. A voice can be imitated. A face can be animated. A scene that never existed can be made credible. A document can be generated in the exact style of an institution. A conversation can be reconstructed with enough fluency to fool an untrained observer.

The problem is no longer simply that falsehoods exist. The problem is that falsehood can now be produced at industrial scale, tailored to the target, and generated almost instantly.

When the cost of manufacturing decreases, value shifts towards authentication. This is a basic economic law.

Abundance destroys one form of scarcity and creates another. In the synthetic world, scarcity will no longer necessarily be the image, the text, or the voice. It will be the proof of their origin.

Artificial intelligence can help examine the evidence; it must not itself become the evidence of what it claims.

This transformation affects almost all institutions in society. The media relies on verification. Justice relies on evidence. History relies on archives. The Art Market relies on authenticity and provenance. Science relies on the traceability of results. Democracy relies on a minimum of shareable facts.

If the distinction between document and simulation becomes more difficult, all of these architectures must evolve. Therefore, this is not a problem reserved for digital technicians.

This is a potential epistemological crisis: how do we know that something has happened?

This question seems almost primitive. Yet it returns to the forefront with extraordinary force. We had forgotten that a huge part of civilization rests on conventions of trust.

When I read a newspaper, I do not personally check every photograph. I trust a chain: journalist, agency, editor, publisher. When a judge examines a piece of evidence, they ask where it came from, under what conditions it was seized, and whether it has been altered. When a collector looks at a work of art, they consider the signature, the provenance, the catalogues, and the exhibitions.

In every field, trust is never absolute. It is built upon a foundation. It is precisely this word that becomes central.

Provenance is not just the origin of an object. It is the chain that allows us to reconstruct its history.

Who produced it?

When?

Or?

Who transmitted it?

What transformations?

This genealogy provides the evidentiary value. In the Art Market, we have always known that an object can be perfectly visually appealing yet problematic if its provenance is inconsistent. The eye alone is not enough. The documentation is essential. AI is now extending this truth to the entire informational world.

An image may look perfect. The question now is: what device produced it? When? Under what conditions? Has it been modified? By whom? Provenance becomes a technical chain.

Photography had long enjoyed an implicit privilege of reality. We knew it could be staged or retouched, but it generally remained linked to the existence of an event in front of a lens. This link is now technically broken. A photograph can be produced without an event.

This represents a huge anthropological shift.

The image ceases to be necessarily a luminous trace of reality and sometimes becomes a pure statistical construct. This obviously does not destroy photography as a medium. But it does destroy a presumption. We must learn to look differently.

For a long time, the image almost silently said: this was in front of the lens. Now, it can only say: this looks like something that could have been in front of a lens.

The difference is striking. And this difference can be exploited politically.

The first reaction is naturally to fear that the fakes will be believed. But a second danger is perhaps even more profound: that the real documents will be denied. When technology makes it possible to fabricate almost any scene, an official confronted with genuine evidence can respond: it is a deepfake. This is what some have called the liar's dividend.

The general possibility of falsehood weakens the strength of truth.

This is why the crisis of evidence is not limited to detecting forgeries. Even an extremely sophisticated detector will not suffice if society loses its trust in all traces. The danger lies in the blurring of distinctions. Everything becomes potentially false.

And when a population no longer knows what to believe, it may end up believing what best corresponds to its identity or its emotions. Factual truth gives way to belonging.

This is where the synthetic world directly intersects with cognitive warfare. The most effective strategy is not always to make people believe a specific story. It can be to destroy the very idea that a story can be verified. To saturate the information. To contradict it. To produce ten versions of the same event. In the end, the citizen no longer chooses based on evidence but on personal preference.

It is a form of epistemic nihilism.

Nothing is true, therefore everything is possible. A democracy cannot function like this for long. It naturally accepts disagreements over interpretation. It cannot function without any common ground of facts.

We can debate the meaning of a demonstration. But we must be able to establish that it took place, approximately how many people were present, and what documented events occurred. Without this foundation, the debate dissolves.

This is why the new architecture of trust should be less concerned with producing a central truth than with making chains of evidence verifiable.

I am deeply wary of any institution that claims to become the universal arbiter of truth. History demonstrates the danger of such a concentration of power. Good governance, therefore, is not a ministry of truth. It is a plurality of mechanisms that allow for the verification of origins and the comparison of sources.

The distinction is that of authority and provenance.

An authority says: believe me. A source says: here's how you can rebuild the chain.

This second approach is much more compatible with an open society. It makes trust verifiable rather than simply required. This is also what a large database does when it is properly constructed. It does not just display a result. It allows you to trace it back to the source, the date, the actor.

This is why structured data has greater power than a simple assertion. It contains its genealogy.

This idea has permeated Artprice. A sale result is not valuable simply because a number appears on the screen. It has value because it can be linked to an Auction House, a date, a lot, a work of art, a currency, a catalogue. It is this chain that builds trust.

The synthetic age will extend this need to almost all information. We will probably enter a general economy of provenance.

Devices will be able to cryptographically sign certain content as soon as it is captured. A photograph could be accompanied by technical information indicating that it comes from an identified device, that it was taken on a specific date, and that it has undergone a certain type of modification.

This architecture will never make forgery impossible. No human system is inviolable. But it will increase the cost of falsification and make verification easier. That is the goal.

Security is not about eliminating all fraud. It is about making fraud difficult and detectable enough.

It is a classic approach. Banknotes can be counterfeited. Security features are added. Passports too. The digital world will have to do the same for content. A kind of watermark indicating origin.

But we must be very careful with words. A watermark embedded in synthetic content indicates that it was generated by a specific system, but it can be removed, bypassed, or lost during transformations. The most robust architecture will likely be one that combines several layers: signature at capture, modification chain, platform attestation, and independent verification.

More redundancy.

A single piece of evidence can be falsified. Multiple independent signals increase confidence. The same principle applies to military and scientific systems.

However, every chain of trust has a limit: it starts somewhere. You have to trust a device, an institution, a certification authority.

We can never completely eliminate trust. We can only distribute it and make it more auditable.

This is an important philosophical lesson. The technological fantasy of a purely mathematical truth is an illusion. Even a cryptographic signature guarantees that certain content was signed with a certain key. It does not automatically guarantee that the scene depicted is true. An authenticated camera can film a staged event.

Provenance does not replace investigation. It strengthens it.

Art presents a similar case. A perfect provenance is not always enough to authenticate a work if the chain itself is fraudulent or if the attribution remains disputed. Expertise, scholarship, and archival research must be combined.

The truth is often a body of evidence.

The synthetic age must learn to think in terms of data sets rather than single documents. A single video may no longer suffice in certain critical contexts. We look at metadata, other cameras, witnesses, satellite imagery, and the timeline.

AI therefore becomes simultaneously a machine for producing the false and a machine for detecting it.

The same duality reappears.

The same technology produces both the weapon and the detector. This will create a perpetual race. Generators are improving, detectors are improving. But I do not believe we can build our trusted architecture solely on automatic detection.

The problem is similar to that of antivirus software: detection often comes after the adversarial innovation. A positive layer of origin is necessary.

Do not just ask: Is this content false?

But: what is its original channel?

It is much more robust. The difference is profound.

Detection looks for a flaw. Provenance builds an identity.

The passport provides a simple analogy. A person is not recognized solely by looking for signs of a false identity. They are given a document, biometric data, and an administrative process. This does not eliminate fraud, but it builds trust.

We may need to do something similar for information of high democratic, legal or historical value.

But beware of the opposite risk: building a world where only officially authenticated content is considered legitimate. This would be extremely dangerous. History, journalism, and whistleblowers often rely on anonymous documents, unverified phones, and clandestine sources.

An architecture of origin must never become an architecture of censorship.

We must retain the possibility that a document without certification could be true. Its initial level of trust will simply be different and will require more verification. Again, it is about granularity.

The synthetic world forces us to abandon binary categories.

Authenticated does not automatically mean true. Unauthenticated does not automatically mean false. They are elements within a broader evaluation.

Intelligence lies precisely in managing this uncertainty.

This is where journalistic culture becomes extraordinarily valuable. A good journalist knows that a single source is not enough for a serious story. They seek independent confirmation. They know the difference between a source close to the case and documentary evidence.

The provenance of information becomes strategic.

This ancient discipline suddenly becomes a technology of democratic survival.

We are therefore entering a dangerous economic asymmetry: the fake can be produced quickly while the real always requires investigation.

It is a battle between speed and method. Digital platforms have accustomed audiences to immediate reactions. But verification requires time. We see this idea of civilizational slowdown again. Sometimes we have to accept not knowing immediately.

This phrase could become a major media virtue: we cannot yet authenticate this document.

In the attention economy, this caution seems weak in the face of an account that makes a definitive statement. But it is the only serious method. Democracy must therefore relearn to value declared uncertainty.

The journalist who says "we are checking" is perhaps more reliable than the one who immediately makes a definitive statement. This is a difficult cultural shift. Social media often rewards certainty and emotion. Evidence requires nuance.

AI can help media outlets speed up certain fact-checking processes without eliminating this discipline. It can compare an image to archives, analyze geolocation, and search for previous publications. But the decision to publish must remain linked to editorial responsibility.

Here again, a human signature is required. AI can assist. The media takes responsibility.

Trust becomes capital again.

It is almost a revenge of the slow institutions.

For a time, the internet gave the impression that intermediaries would disappear. Everyone could publish. It was an extraordinary democratization. But when everyone can also create something plausible, trusted intermediaries regain their function.

Not necessarily the same as before. New players will emerge: verification platforms, origin registries, independent authorities. But intermediation returns. It is an almost cyclical pattern.

When information was scarce, the intermediary distributed it. When information becomes abundant, the intermediary filters and authenticates it.

The problem, then, is to avoid concentrating this power excessively. A single, universal auditor would be dangerous. We need an ecology of trust. Multiple institutions, multiple methods, the possibility of challenging decisions. Pluralism again. Democratic trust must never become a monopoly.

This pluralistic architecture seems to me extremely close to the functioning of a healthy Art Market. Attributing a work can involve historians, catalogues raisonnés, scientific analyses, provenance research, and experts. No single element is always definitive. A solid attribution often arises from convergence.

It is precisely this beam model that should be generalized.

The digital world had become accustomed to binary instantaneity: true or false. Reality is often more complex. Some images are authentic but taken out of context. This is probably one of the most effective forms of manipulation.

We do not even need to create a fake. We take a real photograph of a past event and claim it is from today. Technically, the image is authentic. Semantically, the usage is false.

That is why technical authentication is never enough. You need the content history.

When did it first appear? In what context? That is precisely the origin. Evidence has a temporal dimension.

And AI can traverse this timeframe very quickly. It can find that a video supposedly filmed yesterday was already circulating three years earlier in another country. It is a fantastic tool. But the archives must exist and be accessible. We are back to the issue of memory again.

A society without archives is extremely vulnerable to false context.

If we do not preserve records, we cannot make comparisons. This gives digital libraries, press archives, and image databases enormous strategic value. Memory becomes a defense mechanism.

This formula seems very powerful to me: memory is an infrastructure of cognitive security.

A society that preserves its archives can better recognize reuse, manipulation, and rewriting. This deeply resonates with my obsession with preservation. I have always believed that an old document can find an unexpected purpose decades later. AI confirms this.

The archive is not only a look toward the past. It becomes an instrument for verifying the present.

It is a transformation of its function. And that is exactly what I experienced with the Art Market archives. A hundred-year-old catalogue can authenticate a work today. A mention, a photograph, a dimension. The past becomes proof of the present.

This structure is becoming widespread. A ten-year-old web archive can demonstrate that a claim presented as new already existed. An old photograph can verify a building. Memory becomes a witness.

Archives must therefore be protected against manipulation. If a company begins to rely on digital archives to verify content, the integrity of those archives becomes critical.

Who can modify them? Are there multiple copies? Are the versions time-stamped?

More redundancy. A single archive controlled by a single entity is vulnerable. Heritage institutions, universities, and media outlets could maintain independent copies. This resembles the distributed internet. Once again, we see the logic of the network.

Documentary truth gains robustness when it has no single guardian.

The judge will then have to distinguish the visual appearance from the technical chain that produced it.

The judge will no longer be able simply to look at the document. The judge will have to examine its technical provenance.

Who seized it?

When ?

Which hash?

Which copy?

What treatments?

Criminal procedure is entering the cryptographic age. But we must ensure that technical sophistication does not render justice inaccessible to citizens and lawyers. If only a few experts are capable of understanding the evidence, we create a new asymmetry.

The defense must be able to challenge the methodology. This is a fundamental principle of adversarial proceedings. Algorithmic or digital evidence that cannot be examined by the opposing party poses a democratic problem. Trade secrets cannot become an absolute obstacle to the defense.

Justice requires the capacity to audit.

If a system claims that a voice is 98 percent authentic, you need to understand the method, the error rate, and the reference data. The score alone is still not sufficient. Probability does not become certainty just because it has two decimal places.

This is a scientific illusion that AI could amplify.

The law must therefore learn to distinguish evidence from probability.

The potential for errors are numerous: false positives, data bias, models poorly suited to the context. The justice system will likely need to develop specific standards for evidence generated or analyzed by AI. Not to reject it outright, but to calibrate its weight.

Just as DNA gradually found its place with procedures and limitations, every emerging technology goes through a phase of institutional learning. AI will be no different.

But it adds a unique feature: it can produce the disputed content itself. The judge will sometimes have to determine not only whether a document is authentic, but also whether it was artificially generated.

This expertise will be a constantly evolving field. Detection tools can become obsolete. That is why source evidence remains superior when available. If a police camera cryptographically signs a video upon capture, the chain of custody is stronger than if one tries months later to determine whether the file is synthetic.

This does not resolve questions of context or operator behavior. But it does reduce some uncertainties.

We therefore see a general principle emerging: to authenticate as early as possible in the chain.

It is almost an industrial rule. Traceability is more reliable when it begins at the production stage. In art, this is precisely why a living artist's certificate or the studio archive can become so important. The later the reconstruction, the more doubt arises.

Synthetic generation could profoundly transform the notion of artistic signature. A work produced with AI can have a human author who defined the intention, selected, and reworked it. The debate about authorship is complex and cannot be reduced to the question of whether a machine participated.

Artists have always used tools, workshops, and assistants. What's changing is the level of autonomy in the process.

But the Art Market will above all have to answer a question of traceability: what is the genealogy of the work? What models, what prompts, what interventions, what versions?

The archive therefore forms an integral part of the creation process. This constitutes a new form of archive of creation.

For an art historian, this will be extraordinary material. We will sometimes be able to reconstruct the stages of production with unprecedented precision. But we will have to choose what should be preserved.

All the prompts? All the versions? The creation process already generates enormous amounts of data. A complete archive becomes impossible. Selection will be necessary. This brings us back to a problem I have always known: keeping a lot, but structuring it so that the memory remains usable.

Unorganized abundance is a form of amnesia.

This statement seems fundamental to me for the synthetic age. We can record almost anything and then never retrieve it again. Memory is not quantity. It is architecture. This is precisely what AI can improve, but only if metadata and provenance exist.

The Abode of Chaos has always been a place where matter and trace coexist.

That is why it seems to me to have anticipated the current crisis of evidence.

Each portrait is linked to an event, an era, a geopolitical memory. The wall functions almost like a distributed physical archive. Even if digital technology were to disappear, some of the traces would remain embodied in matter.

This physical dimension takes on greater importance when synthetic production becomes easier.

A dated printed document, a silver gelatin negative, an engraving, a physical inscription possesses certain specific resistances. They too can be falsified, of course. But the physical medium adds another layer. It is almost a revenge of the analog.

We might see a re-evaluation of certain physical media as anchors of trust. This does not mean returning to paper for everything. But preserving certain important records in formats that are difficult to alter could become relevant.

A company entirely dependent on editable files is vulnerable.

Again, the redundancy between digital and physical. This hybridization seems very healthy to me. The most important archives can exist in multiple forms. The digital world does not abolish matter. It sometimes enhances its value precisely when the immaterial becomes too easy to manipulate.

Physical artworks possess this power. Their aging, their materiality, their imperfections are all part of their presence. A perfect synthetic image can visually imitate a painting, but it does not automatically possess the object's material history. This difference may become even more significant in the future.

The aura, to use Benjamin's term, could change in nature.

In a world where reproducibility is becoming limitless and generative, the historical uniqueness of an object can gain value. Not because digital technology is aesthetically inferior, but because the object carries an embodied provenance. It has been there. It has aged. It has been touched, moved, preserved. It is a physical temporality.

Here I find all the elements of the Abode of Chaos: rust, steel, impacts, pigments, scars. These are not textures applied to produce an effect. They are material events. The rain has done its work. Time has done its work. The material bears witness.

This difference between a simulated ruin and a ruin actually affected by time may become increasingly culturally perceptible.

The synthetic world can paradoxically increase our desire for reality.

This is an important hypothesis. When perfect images become infinite, physical presence, chance, and imperfection can become more precious. We have already observed a return to analog in certain artistic practices. This is not simply nostalgia. It is sometimes a search for resistance.

Digital technology makes many things smooth and reversible. Matter produces irreversibility.

A sculpture that is cut, welded, burned represents a decision. It is not simply a matter of pressing Ctrl-Z. This irreversibility lends it gravity. It connects to responsibility.

In the synthetic world, a thousand versions can be generated before a choice is made. The cost of experimentation plummets. This is extraordinary for creativity. But it can also diminish the value of the decisive gesture. The artist may have to find that decision elsewhere: in the selection, the embodiment, the exhibition.

Creation does not disappear. Its location shifts.

It is exactly like work. When a machine automates a function, scarcity shifts to another. In synthetic art, image production becomes abundant. What becomes rare can be vision, context, signature, the ability to construct a work over time.

An image does not make a work of art.

This is a distinction that art history knows very well. A work of art exists within a trajectory, a body of work, an intention, and a reception. AI does not eliminate any of these dimensions. It simply makes visual production less rare.

This may even clarify what art has never been: mere technical virtuosity.

Photography had already shifted this boundary. Duchamp had shattered it. Artificial intelligence continues this story. It forces us to ask once again: where is the artistic act?

It is a fascinating question, but it is not the heart of this chapter. The heart remains the evidence. Yet art offers an extraordinary laboratory because it has long worked with authenticity, fakery, copies, and appropriation.

The Art Market already has procedures that the rest of society may rediscover: catalogues raisonnés, certificates, comparative expertise, scientific analysis, and provenance chains. This knowledge could become surprisingly relevant to the digital world.

A synthetic civilization will need historians of provenance.

It is almost a new profession. Not just fact-checkers who examine a claim, but specialists capable of reconstructing the genealogy of a piece of content.

Where did this video come from?

What versions exist?

Who published it first?

Which accounts shared it?

Which model was possibly used?

AI can automate much of this work, but interpretation will remain important. We will likely see the emergence of visible provenance graphs. Content could have a kind of passport showing its chain of transformations. This representation would be extremely powerful for users.

Rather than a simple "true" or "false" label, we would see: captured by this device, modified with this software, cropped, published by this organization. This restores the complexity.

This is exactly the kind of architecture I prefer. Not imposing a single conclusion when you can show the chain of reasoning. The citizen can then exercise their own judgment.

This obviously requires a certain level of literacy. Media education must therefore evolve. For a long time, the focus was on verifying sources and cross-referencing news outlets. Now, it will be necessary to add an understanding of synthetic content, metadata, and recommendation mechanisms.

But we mustn't turn every citizen into a forensic expert. The architecture must make verification accessible.

A good trust system does not rely solely on the heroic vigilance of the user. It is exactly like cybersecurity. We train people against phishing, but we also build filters and safeguards. The responsibility is shared.

Platforms therefore play a crucial role. They can detect certain synthetic content, display its origin, and slow the spread of unverified material in critical situations. However, this intervention must be regulated to prevent arbitrary censorship.

Here again, procedures matter.

Who can challenge a label? How transparent is the method? How long does it take?

The appeal process always returns. Content wrongly classified as synthetic can cause enormous harm, especially for a media outlet or a whistleblower. Therefore, an appeals architecture is necessary. Appeal by design, once again.

We are seeing the formation of a true constitution of digital trust: origin, pluralism, audit, recourse.

The same principles keep recurring because they are fundamental. The proof crisis will not be solved by a single miracle technology. It requires an ecosystem. That is why I talk about a new architecture of trust.

An architecture has several layers. At the capture level: signatures and metadata. At the platform level: identification of transformations. At the media level: editorial verification. At the legal level: standards of evidence. At the educational level: literacy. At the political level: transparency. No single layer is sufficient on its own.

Robustness comes from combination. It is still the stack: no cognitive sovereignty exists without the material, energy, computational, and documentary chain that makes it possible. Trust, too, becomes vertical.

And this architecture must be international. A video crosses borders in seconds. Therefore, standards of origin must be interoperable. Otherwise, each country will have its own system, and global verification will become difficult. This creates a new geopolitical challenge related to standards.

Who defines the standards of digital authenticity?

The United States, China, and Europe may have different philosophies. An overly centralized architecture can create opportunities for political control. An overly open architecture can be more vulnerable to abuse. It will be necessary to find sufficiently neutral common protocols.

It is exactly like the internet. The technical layers may be shared, but the regulations differ. Perhaps a global standard will allow us to declare that content has been modified without specifying who has the right to publish it.

Separating information about the source from the political decision regarding the content would be a healthy distinction.

The protocol states: this file has this history. The company then decides what to do with it. This philosophy is similar to open standards. It reduces the risk of the technical layer becoming a tool for censorship.

But no system is absolutely neutral. Even the choice of metadata reflects priorities. Therefore, pluralistic governance of standards will be necessary. International organizations, industry, media, and civil society will have to participate. It is a slow process. But again, some delays are useful.

A global standard imposes long-term consequences. There needs to be a degree of contradiction.

We find ourselves back with the ever-changing permanence: a protocol stable enough to generate trust, yet scalable enough to keep pace with new attacks. Maintenance will be continuous. Because the fake will adapt.

Malicious actors will learn to exploit the new systems. They will steal keys, compromise devices, and create fraudulent origin chains. It is inevitable. Absolute security does not exist. That does not make the effort pointless.

It is exactly like the law. Crime exists despite the Penal Code. We do not conclude that the law should be abolished. We build procedures, we improve protections.

Trust will always be probabilistic. We must accept that.

Perhaps one of the biggest cultural shifts will be to stop treating content as immediately true or false and to develop levels of trust. This brings everyday life closer to the scientific method.

A statement can be well-established, plausible, uncertain, or refuted. This language is richer.

AI could actually help democratize it. Instead of responding with artificial certainty, it could indicate: this information is supported by several independent sources, this information is confirmed by only one source, this information is disputed. That would be extraordinarily useful.

But this requires that the systems be designed for traceability.

An AI that produces an answer without a visible source contributes to the problem of proof. It may be correct, but the citizen does not know why. That is why source information must also be factored into AI assistants.

Where does the information come from? How up-to-date is it? Which sources differ?

This architecture transforms AI from a new oracle into an interface to the evidence. This is precisely the direction that seems democratically sound to me. AI should not replace the source. It should help to navigate it.

This distinction resonates deeply with our own dialogue. The value of AI lies not only in producing a fluent sentence, but also in connecting elements, finding information, and comparing them, while leaving the final decision to the human. A response without a source can be elegant but fragile. A response grounded memory becomes more robust.

This brings me back to the Abode of Chaos as a living, breathing entity. It is powerful because it is not a generated, memoryless stream. It possesses twenty-five years of layers, traces, and transformations. Each new gesture is inscribed on an already charged surface. It is the exact opposite of an amnesiac, synthetic world where each image appears without history.

The Abode of Chaos is a materialized origin.

A scar is not just an image of a scar. It is the trace of a past action. This is perhaps one of the reasons why it becomes even more relevant in the age of AI. It affirms that memory is tangible.

The synthetic world can produce any appearance. It cannot retroactively produce twenty-five years of lived causality without constructing an extremely complex fiction. History remains a form of resistance.

I propose here another hypothesis: temporal depth is a form of authenticity.

A person, a work of art, an institution accumulates coherent traces over years. It becomes more difficult to simulate this totality than to simulate a single moment. The forgery excels in the fragment. Provenance reconstructs duration.

Therefore, temporality can become one of the best antidotes to the synthetic.

Do not just ask: does this image look real?

But: how does it fit into a chronology? Are there any compatible earlier traces?

This method is exactly that of the historian. The age of AI may well make everyone a little bit of a historian. We will have to learn how to research genealogy.

It is a beautiful irony. A futuristic technology brings us back to a very old discipline: source criticism.

Historians have known for centuries that a document must be contextualized, compared, and dated. Digital technology does not abolish this method; it makes it universal.

This is probably one of the major lessons of this chapter. Faced with artificial intelligence, we do not need to abandon all the old methods. On the contrary. We sometimes need to reactivate them with new tools. Source criticism, legal debate, artistic provenance, the scientific method — all these traditions were already technologies of trust. AI does not make them obsolete. It increases their importance. Only their instruments change.

Here again is the mutant permanence.

Civilization does not start from scratch. It adapts its institutions to the new medium. But the synthetic world introduces an additional question that old methods did not have to confront on this scale: the personalization of the fake.

The same event can be recounted differently to each person. AI can generate a video in which a politician appears to say exactly what will trigger outrage in a specific demographic. This fragments the evidence. Two citizens could receive two different fake videos and believe they are talking about the same event. How can we build a public debate in this situation?

Public provenance then takes on even greater importance. The media and institutions may need to maintain accessible repositories of authentic statements, speeches, and official documents—a kind of verifiable civic memory.

This already exists to some extent, but AI is increasing the need. A public official could cryptographically sign their official statements. Again, this would not make imitation impossible. But citizens could verify: does this video correspond to an authenticated statement? This is a concrete example of a new trust architecture.

But we must maintain space for independent journalism. An unsigned statement can be authentic. A clandestine video can reveal the truth. We must never confuse official with true. This is crucial.

Provenance means origin, not automatic legitimacy.

An official document can lie. An anonymous document can reveal the truth. That is why pluralism remains. The source helps us know what we are looking at, not necessarily whether the proposition it contains is true. This nuance must be taught everywhere.

Otherwise, we risk replacing synthetic chaos with a centralized authority of authenticity. This would be a symmetrical error.

Good architecture is that which increases the information available for judgment without eliminating judgment.

That is exactly my philosophy of AI: augment, do not abdicate.

We can then imagine a kind of digital trust layer where each important piece of content has several attributes: declared origin, verified origin, known transformations, corroborating sources, and disputes. The citizen would see not an absolute green badge, but a map.

This is a more mature way of representing knowledge. It directly relates to the representation of uncertainty I discussed in the context of cognitive warfare. The interface should not obscure complexity when it is relevant.

An honest trust architecture must be able to say: we know this with a high degree of confidence, but this element remains uncertain.

That is precisely what science does. The problem is that media and political discourse often favors certainty. Therefore, we must also change the culture. AI will not solve a society that rejects uncertainty. It can only provide the tools.

This is still the rule: technology amplifies the institution.

A society that values verification can use AI to improve verification. A society that values confirmation can use it to produce more content that aligns with its beliefs. The crisis of evidence is therefore also a crisis of the desire for truth.

This psychological dimension is essential. People do not always share content because they believe it to be true. They may share it because it expresses what they would like to be true. Emotion sometimes precedes verification. Generative AI can exploit this with formidable efficiency.

A purely technical approach to provenance will therefore not suffice. An education in the desire for proof is necessary.

This expression may sound strange, but I believe it is true. A society must value changing its mind when contradictory evidence emerges. It is a rare intellectual virtue. The opposite is dogmatism.

AI could exacerbate dogmatism by constantly providing an argument for everyone. With enough data generation, it is possible to find a text confirming almost any intuition. That is why the tool must be designed to also search for objections.

The Blind Spot again.

A trust architecture should present not only the elements that confirm a hypothesis, but also those that weaken it. This is precisely the adversarial method.

We see that all the threads of this section converge: origin, contradiction, pluralism, uncertainty, responsibility. These are the structures of truth in a complex society.

AI makes them visible because it threatens to short-circuit them. But it can also strengthen them. It can sift through more sources, detect contradictions, and reconstruct a timeline.

That is why I reject catastrophism. The synthetic world is not necessarily the end of truth. It can be the beginning of a more sophisticated proof architecture.

The interface should not conceal complexity when it matters.

This will require effort. But societies have already weathered similar crises with each new medium. Printing multiplied texts, including false ones. Photography created new possibilities for manipulation. Radio, television, the internet. Each time, institutions and audiences learned new methods.

AI accelerates and amplifies the challenge, but it does not place us out of history.

This continuity is reassuring because it reminds us that civilization knows how to mutate. Once again: only the mutants survive.

The problem is not the emergence of synthetics. The problem lies in maintaining institutions designed for a world where the cost of counterfeiting was much higher.

The transformation must be rapid. Courts, media, schools, and government agencies must adapt their procedures. But the core remains stable: evidence, adversarial proceedings, and accountability. This is precisely the essence of enduring change.

The synthetic world could even produce a new renaissance of materiality and presence. When we are no longer certain of an image, physically attending an event can regain its value. The concert, the museum, the conference, the meeting.

Presence becomes existential proof.

“I was there.” Of course, the testimony could be false. But the embodied experience acquires a different meaning in a universe mediated by synthetic content.

Perhaps physical places will become sanctuaries of trust.

This immediately brings me back to the Abode of Chaos. You can see millions of images of the Abode of Chaos. But physically entering the 9,000 square meters, feeling the material, seeing the scale, the scars, is an incomparable experience.

Reality resists reproduction.

This is precisely why the physical museum does not disappear with digitization. On the contrary, the more powerful the digital technology, the more valuable the physical space can become. This is a paradox we are already observing. Global reproduction increases the visibility of the unique object.

AI could take this logic much further. We could generate an infinite number of imaginary museums, impossible sculptures. This would not necessarily eliminate our desire to be in front of something that truly exists. Perhaps even the opposite.

Presence will become a cognitive rarity. And rarity produces value.

This is how the synthetic world can restore authenticity without reviving a naive nostalgia for the old world. The digital and the physical do not cancel each other out. They redefine each other.

AI can enrich the visit, link the works, and display the archives. The physical world brings presence.

This is where Artprice and the Abode of Chaos meet again in my thought: a global database on one side, a total physical work on the other. Digital memory and a plastic body. Artificial intelligence becomes a nervous system between the two.

This triptych now takes on an even greater dimension. The base enables origin. AI enables relationship. The Abode of Chaos provides embodiment.

It is almost a complete architecture of trust: memory, intelligence, matter.

I had not necessarily formulated it that way at the beginning of this adventure, but it is becoming increasingly clear. And this architecture now raises an even deeper question.

Even if we manage to authenticate images, voices, and documents, there will still remain a form of synthetic that cannot be dealt with solely by provenance: identity itself.

Artificial agents can have a voice, a face, and conversational memory. They can present themselves as assistants, advisors, or digital companions.

At what point do we need to know that we are talking to a machine?

Does an AI always have to identify itself as such?

What is identity in a world where the same person can delegate part of their presence to agents who speak on their behalf?

We then enter into an even more dizzying question. The proof no longer concerns only the content. It concerns the interlocutor.

Who is speaking to me?

Is it Thierry Ehrmann, his authorized AI agent, a model imitating Thierry Ehrmann, or a fake?

This distinction will become crucial for commerce, politics, and personal life. Identity itself will need to have a provenance. We will see the return of the signature, the mandate, the authorization.

An artificial agent may be able to negotiate on behalf of a person or a company. It will be necessary to define precisely what actions it is authorized to perform. This resembles the law of agency, but on a new scale and at a new speed.

An agent can be involved in a very large number of transactions simultaneously. It is not human identity itself that is being distributed, but rather actions carried out in its name. This difference makes mandates and traceability even more necessary.

This idea is fascinating and dangerous. It could dramatically increase our capacity for action. But it can also fragment responsibility.

If my agent promises something that I would not have personally phrased that way, am I bound?

We find the chain of responsibility again.

After the crisis of proof comes the crisis of the interlocutor.

We will no longer simply ask:

Is this image real?

We will also ask:

Who am I actually talking to?

And this question could transform commerce, law, and social life as profoundly as the emergence of AI itself.