What No One Knew How to Protect

Constitutions describe the rules of the game. They establish rights, organize powers, and provide remedies. But they say nothing about what makes these rules acceptable — the trust that citizens place in institutions, the legitimacy that distinguishes an accepted decision from a mere exercise of force, the authority that allows certain voices to be heard as trustworthy. Pierre Rosanvallon calls these prerequisites for democracy the invisible institutions. And he shows, in a work published by Seuil in 2024, that they are crumbling without anyone really knowing how to defend them.

The Essentials

• Rosanvallon identifies three unwritten pillars of democracy: trust, authority, and legitimacy — none of which benefits from formal institutional protection
• Trust in institutions has undergone a profound collapse in Western democracies; in France, according to the CEVIPOF Political Trust Barometer (since 2009), trust in the National Assembly reached 24% in 2024, and trust in the government was approximately 23% in 2025
• The work proposes a complete theory of democracy: formal procedures are insufficient if the informal substrates that make them function disappear
• European AI regulation (AI Act, DSA) directly touches the algorithmic production of common sense — where the defense of these invisible pillars is concretely at stake

Pierre Rosanvallon

Pierre Rosanvallon teaches at the Collège de France, where he has held the chair of modern and contemporary history of the political since 2001. A historian and political philosopher, he has constructed over four decades a coherent body of work: understanding how democracies produce their own legitimacy and how they lose it. His work on “unfinished democracy,” “counter-democracy,” and “the elusive people” form a corpus that makes him one of the most important thinkers of contemporary democratic crises. The Invisible Institutions is his synthesis work: after mapping formal institutions, Rosanvallon descends to their foundations.

The Central Thesis: Democracy Rests on What It Failed to Formalize

Rosanvallon’s proposition is both simple and troubling. Liberal democracies have spent two centuries building formal institutions — constitutions, separation of powers, fundamental rights, electoral procedures. This formalization work produced systems that are remarkably solid on paper. But these systems rest on assumptions that constitutions never formalize: the conviction that facts exist and are shared, that experts can be believed on certain subjects, that institutions deserve initial trust before having to earn it case by case.

Rosanvallon calls these assumptions the “invisible institutions.” Invisible does not mean non-existent: these pillars are as real as the articles of a constitution. But they are written nowhere, and no one is mandated to defend them. “Trust is to democracy what credit is to the economy,” he writes: it allows the system to function in advance of proof, to make decisions without having to justify them completely at every moment. When credit collapses, an economy stops. When trust collapses, a democracy spins idly.

The originality of the argument lies in its structure. Rosanvallon does not say that formal institutions are insufficient — he says they are conditional. They function only if the invisible institutions hold. And these latter, no one has ever conceived of them as collective goods to protect.

Three Pillars, Three Crises

The work distinguishes three main invisible institutions — trust, authority, and legitimacy — and for each, Rosanvallon reconstructs the ongoing crisis.

Trust, first. It is not naïveté or blind deference. It is a probabilistic judgment: I trust an institution because I have reasons to believe it acts in the general interest, even if I cannot verify each decision. This trust has collapsed in most Western democracies over the past thirty years. In France, the CEVIPOF Political Trust Barometer — which has measured these developments since 2009 — documents this collapse: trust in the National Assembly reached 24% in 2024, and trust in the government was approximately 23% in 2025. The European Union is experiencing a comparable trajectory.

What Rosanvallon emphasizes is that this collapse is not merely a reflection of bad political decisions. It signals a rupture in the very form of the bond between those who govern and the governed: citizens no longer believe that institutions play the role they claim to play. It is a crisis of the regime of institutional truth, not merely of their performance.

Authority, next. Rosanvallon carefully distinguishes authority from power. Power can be decreed, imposed, transmitted. Authority is recognized: it is accorded to someone or something by virtue of recognized competence or integrity. Scientific experts, constitutional judges, system arbiters have an authority that exceeds their formal mandate. This authority is an invisible institution: it is written nowhere, but it makes the system function.

The crisis of scientific authority rendered this dimension concrete. The Covid-19 pandemic showed how formally solid public health systems could find themselves defenseless facing a challenge to the epistemic authority of scientists. The institutions were there. The authority that allowed them to be listened to had partially decomposed.

Legitimacy, finally. By this, Rosanvallon means the set of shared reference points that allow citizens to recognize that a decision, even if contested, was made according to rules and values they accept as foundational. This legitimacy is a condition of possibility for democratic deliberation. Without it, public debate no longer produces decision: it produces polarization.

This is where the book articulates with digital transformation in a way that Rosanvallon does not avoid: recommendation algorithms are not neutral with respect to common sense. They produce it. By choosing what each individual sees, reads, and hears, they manufacture competing realities sealed off from one another.

AI Regulation as a Politics of Invisible Institutions

One of the most stimulating contributions of the book is its ability to reread technological regulation from a new angle. If algorithms are machines for producing common sense — or destroying it — then the rules that govern their functioning are, in Rosanvallon’s sense, policies for protecting invisible institutions.

The European AI Act, whose regulation EU 2024/1689 was published in the Official Journal of the European Union on July 12, 2024, and entered into force on August 1, 2024, and the Digital Services Act (DSA), applicable since 2023, can be read differently through this lens. The transparency obligations on algorithmic systems, the rules on deceptive content, the requirements for pluralism in recommendation systems: all of this is not merely sectoral regulation. It is an attempt, probably the first of this scale, to formalize a protection of the informal substrates of democratic deliberation.

The question remains whether these instruments are up to the task of what they attempt to protect. The DSA imposes on the largest platforms assessments of systemic risks to democracy. But assessment is only valid if the supervisory authorities have the technical and political capacity to make it effective. Yet the European Commission opened proceedings against X/Twitter and Meta in 2024 precisely on these grounds, and the first decisions are slow in coming — a sign that the instrument exists but that its effectiveness remains to be demonstrated. Visible institutions have been created; invisible institutions, meanwhile, cannot be decreed.

The Book’s Blind Spots

Rosanvallon constructs a rigorous theory, but it rests on an assumption he does not interrogate sufficiently: that invisible institutions once existed in pure form and are now in decline. History is more ambiguous. Trust in institutions in the 1960s-1970s, high in opinion surveys, was also trust in systems that were exclusionary — few women in decision-making positions, little representation of working classes, quasi-absence of postcolonial voices. The “common sense” of that era was less a public good than a well-guarded class good.

This does not make the thesis false. It complicates it. The collapse of trust is not univocally a loss: it also accompanies genuine democratization of voices, a pluralization of narratives, a rise in legitimacy of long-excluded groups. The challenge is not to restore the invisible institutions of before, but to build new ones capable of including a society more diverse than it has ever been.

On this point, the perspective of Yascha Mounk — who works on the tension between liberal universalism and the rise of identitarianism — offers a useful counterpoint. Mounk argues that the fragmentation of common sense is as much the product of new political identities as of algorithms: platforms have amplified a tendency toward epistemic tribalization that preexisted in social dynamics. If one follows this reading, algorithmic regulation is necessary but insufficient: it does not address the political roots of the crisis.

There is also a tension between Rosanvallon’s thesis and the vision held by economists like Daron Acemoglu, for whom the degradation of democratic institutions is first and foremost a question of economic capture — elites capturing the rules of the game to consolidate their position. In this framework, “invisible institutions” are not so much common goods eroded by modernity as an ideological veil that long masked capture. Rosanvallon and Acemoglu are not incompatible, but they point to different priorities: one toward restoring the symbolic bond, the other toward redistribution of economic power.

What This Book Changes

The book’s strength is not in proposing a solution — Rosanvallon is lucid about the absence of a simple recipe. It is in displacing the question. For thirty years, the debate on the democratic crisis has revolved around formal institutions: should we have a Sixth Republic, proportional representation, more referendums? These questions are legitimate, but they presuppose that informal substrates are holding. Rosanvallon shows that this is precisely the presupposition that has become false.

This approach directly engages reflection on technology policy. If digital platforms and AI models have become infrastructures for the production of common sense — as the concentration of computing power in few hands makes increasingly evident — then their regulation is no longer a sectoral question. It is a constitutional question in the broad sense: a question about the conditions of possibility for democratic deliberation itself.

The prospective stakes are there. Invisible institutions reconstitute themselves slowly — scientific authority refounds itself on decades of accumulated evidence, trust is reestablished through long sequences of coherent institutional behavior. But their destruction can be rapid. A series of converging data suggests this: according to Eurobarometer, a growing share of Europeans believe that media do not provide reliable information. Among those under 35, trust in democratic institutions is structurally lower than in all previous generations at the same age, according to long-series data from the European Social Survey. This is not a cycle: it is generational sedimentation.

If the generations entering political life over the next ten years approach it with a fundamentally different relationship to shared truth and institutional legitimacy, constitutions will not change. But what they rest on will have changed. This is precisely the asymmetry — the apparent solidity of formal institutions against the growing fragility of their informal foundations — that The Invisible Institutions makes visible.

The open question the book poses without closing it is this: can formal institutions be built to protect informal goods? AI regulation is the first major laboratory of this attempt. Its outcome will tell us much about our capacity to produce democratic progress in a technological environment that makes its conditions more difficult than ever — not impossible, but more difficult than ever.

Pierre Rosanvallon, The Invisible Institutions Seuil, 2024 320 pages

Sources

1. Pierre Rosanvallon, The Invisible Institutions, Seuil, 2024 — review in En attendant Nadeau: https://www.en-attendant-nadeau.fr/2025/05/20/rosanvallon-institutions-invisibles/
2. Political Trust Barometer, CEVIPOF, Sciences Po — annual editions (cevipof.com)
3. Digital Services Act (DSA), European Commission, regulation EU 2022/2065, entered into application in 2023
4. AI Act, regulation EU 2024/1689, Official Journal of the European Union, July 12, 2024
5. Eurobarometer, European Commission — waves 2016 and 2023 on trust in media
6. European Social Survey — long series on institutional trust by age group (europeansocialsurvey.org)
7. Éditions du Seuil - The Invisible Institutions (2024)
8. Collège de France - Biography Rosanvallon
9. EUR-Lex - AI Act regulation EU 2024/1689
10. EUR-Lex - DSA regulation EU 2022/2065
11. Sciences Po CEVIPOF - Political Trust Barometer
12. European Commission - Proceedings against Meta (DSA, May 2024)
13. European Commission - Preliminary Conclusions against X (DSA, July 2024)