Institutions as the Trust Layer
When formats become fakeable, trust migrates to institutions — not better formats
Harari identifies a recurring historical pattern: when a communication technology becomes fakeable or manipulable, trust does not collapse permanently and does not migrate to a 'better' format — it migrates to the institution behind the format. With writing, anyone could put words on paper. Trust migrated to institutions with reputation at stake: The Guardian, The Times, The Wall Street Journal. People don't believe something because it is written — they believe it if it appears on the front page of the New York Times, because they trust the institution, not the paper or the ink.
With video, the format was previously unfakeable, so it was trusted on its format alone. 'If we saw a video, we said ah, this has to be true.' AI deepfakes break that assumption. Harari's prediction, drawn from the historical pattern: the same transition that happened with print will happen with video. 'We are still not used to it. So when we see a video of Donald Trump or Joe Biden, the video still gets to us because we grew up in a time when it was impossible to fake it. But I think very quickly people will realize: you can't trust videos, you can only trust the institutions.'
The democratic stakes: if institutional trust itself collapses — if the institutions that certify content are themselves captured, corrupted, or destroyed — the result is not a neutral epistemic landscape. Harari's argument is that 'if you can't believe anything, this is the ideal for dictators. Democracy works on trust but dictatorship works on terror and fear. You don't need to trust anything in a dictatorship — you fear.' Epistemic collapse is not neutral; it is structurally favorable to authoritarian power.
- When a communication format becomes fakeable, trust migrates to institutions — not to better formats
- Institutional trust is more durable than format trust because it is backed by reputation, accountability, and accumulated track record
- Epistemic collapse is not neutral — it is structurally favorable to authoritarian power that operates through fear rather than trust
- The current transition period (video becoming fakeable) mirrors the historical transition when writing became widely accessible
- The second-order problem — AI becoming the bureaucracy — requires decentralized, cryptographically verifiable certification infrastructure
- Identify the current trust mechanism for any information formatDetermine whether trust in a format is format-based (video, because video was unfakeable) or institution-based (The New York Times, because institutions have accountability). Format-based trust is fragile once fakeability is achieved; institution-based trust persists as long as institutional integrity holds.Pro tipTest: would you believe this content if it appeared on a random website versus on the front page of an established institution you trust? The delta reveals how much of the trust is format-based versus institution-based.
- Apply the historical pattern to predict trust migrationWhen a format becomes fakeable, predict that trust will migrate to institutions rather than collapsing permanently. Identify which institutions are positioned to certify content in the new environment, and assess their integrity and independence. The transition period — before the new trust architecture is established — is the highest-risk period.WarningThe transition period can be exploited by actors who benefit from epistemic chaos. The historical pattern shows the trust architecture eventually stabilizes, but the intervening period can cause significant political and social damage.
- Assess institutional integrity in the current environmentEvaluate whether the institutions that would serve as the trust layer are themselves under threat — from political pressure, financial capture, regulatory attack, or public trust erosion. The institution-as-trust-layer answer only holds if institutions maintain their independence and integrity.Pro tipThe democratic stakes are asymmetric: institutional capture or collapse does not produce a neutral epistemic landscape — it produces conditions structurally favorable to authoritarian power.
- Identify the second-order certification problemWhen AI becomes the bureaucracy that certifies content, ask who certifies the AI certifiers. The institution-as-trust-layer answer requires a further layer of verification that is not capturable by any single actor. Cryptographically verifiable, decentralized verification infrastructure addresses this second-order problem structurally rather than through trust in any particular human institution.WarningCentralized AI certification infrastructure — operated by governments or major platforms — reproduces the capture risk at a deeper level than human institutional capture.
When writing became widely reproducible (printing press, then mass literacy), anyone could put words on paper. The format — written text — became unreliable as a trust signal. Trust migrated to institutions with accountability: established newspapers and publishers whose reputation was staked on their editorial decisions. The institution, not the format, became the trust carrier.
Harari cites Venezuela as the proximate historical example of how democratic institutions can be captured by an elected leader who changes the rules of the game from within. The precondition: once epistemic trust in democratic institutions erodes sufficiently, the population cannot coordinate to resist the capture because they cannot agree on what is true.
This framework emerges from Harari's historical methodology: when facing an apparently novel technological challenge, identify whether history has produced analogous transitions and what the resolution pattern looked like. The writing-to-institution trust migration is well-documented; Harari applies it to video as a predictive tool.
The framework's most important extension is the second-order question Harari names but leaves unresolved: 'Will those bureaucratic institutions be AI?' As AI becomes the bureaucracy that certifies content, the institution-as-trust-layer answer requires a further layer — who certifies the AI certifiers? This is where cryptographically verifiable, decentralized verification infrastructure becomes structurally necessary, not ideologically preferred.