The end of democracy rarely arrives with sirens and flames. More often, it fades quietly—choice by choice, habit by habit, until the rituals remain but the substance has gone.

In their timely paper, Don’t Panic (Yet), Felix Simon and Sacha Altay remind us that the AI apocalypse never arrived in 2024. Despite a frenzy of deepfakes and fears of algorithmic manipulation, the great elections of that year were not decided by chatbots or microtargeted propaganda. The decisive forces were older and more human: politicians who lied, parties who suppressed votes, entrenched inequalities that shaped turnout and trust.

Their conclusion is measured: mass persuasion is hard. Studies show political ads, whether crafted by consultants or large language models, move few votes. People cling to their partisan identities, update beliefs only at the margins, and treat most campaign noise as background static. The public is not gullible. Even misinformation, now turbocharged by generative AI, is limited in reach by attention, trust, and demand.

In this sense, Simon and Altay are right: the panic was misplaced. AI was not the kingmaker of 2024.

But here is the danger: what if reassurance itself is the illusion?

The great risk of AI to democracy does not lie in a single election “hacked” by bots. It lies in the slow erosion of the conditions that make democracy possible. Simon and Altay diagnose panic as a cycle society overreacts to every new medium. Yet what if this is not a panic at all, but an early recognition that AI represents not another medium, but a structural shift?

Democracy depends on informational sovereignty citizens’ capacity to orient themselves in a shared reality. Generative AI now lives inside search engines, social feeds, personal assistants. It does not need to persuade in the crude sense. It reshapes the field of visibility what facts surface, what stories disappear, what worlds seem plausible.

Simon and Altay show that persuasion is weak. But erosion is strong.

• Trust erodes when deepfakes and synthetic voices make truth itself suspect.
• Agency erodes when predictive systems anticipate our preferences and feed them back before we form them.
• Equality erodes when the wealthiest campaigns and nations can afford bespoke algorithmic influence while the rest of the citizenry navigates blind.

In 2024, democracy endured not because AI was harmless, but because old buffers mainstream media, partisan loyalty, civic inertia still held. These reserves are not infinite. They are the borrowed time on which democracy now runs.

So yes: panic may be premature if we define it as fearing that one election will be stolen by machines. But complacency is suicidal if we fail to see how AI, fused with the logics of surveillance capitalism, is hollowing democracy from within.

The question is not whether AI will swing the next vote. The question is whether, by the time we notice, the very meaning of choice will already have been diminished.

Democracy may survive a storm. What it cannot survive is the slow normalization of living inside someone else’s algorithm.