The last time a technological advancement truly upended democratic life, people were still arguing over whether the internet was a passing trend. Now, that argument seems almost charming. The social media era appears to be a mild warm-up for what’s coming.
The concept of autonomous AI agents—systems that can see, plan, remember, and act in various digital environments without much human oversight—is no longer theoretical. They are either here or so close that it is hardly noticeable. Furthermore, the world’s democracies, which are already deteriorating in ways that were unthinkable ten years ago, are utterly unprepared for this moment.
It’s worth taking a moment to consider how tense the situation already is. Currently, just over 25% of people live in electoral or liberal democracies, a decrease from nearly half in 2016. There are currently more than twice as many nations heading toward autocracy as those heading in the opposite direction. These figures don’t come from a dystopian book. They come from political scientists who are monitoring the hollowing out of actual elections, courts, and legislatures. In light of this, the emergence of extremely competent AI agents is more than just a technological development. Institutions that are already failing are being put through a stress test.
This past September, two researchers who have spent time deciphering precisely this issue, Seth Lazar and Mariano-Florentino Cuéllar, published a paper that clearly outlines the stakes without descending into panic. In essence, they contend that AI agents will exacerbate every weakness that democracies already have. Anger among populists is fueled by an acceleration of the loss of white-collar jobs. Concentrating AI infrastructure in a small number of companies gives those companies unheard-of power over the government. Autocrats can now deploy surveillance tools precisely, whereas in the past they had to construct them awkwardly. Additionally, the public sphere, which is already overrun with false information, may become so overwhelming that actual people simply stop believing any of it and withdraw completely.
A specific section of their analysis sticks out. In order to avoid the human checks and balances that typically cause conflict, they describe would-be autocrats using AI-run bureaucracies like an exoskeleton—wearing the state. The image is visceral. It’s also not science fiction. There are already pieces of such machinery in place across several governments.
The fact that the same technology has real democratic potential makes this more difficult to understand. Lazar and Cuéllar take care to keep the potential from being overshadowed by the threat. AI agents could serve as something like cognitive prosthetics, assisting regular people in navigating bureaucratic complexity, spotting manipulation, or organizing around shared causes if they are kept open and decentralized rather than controlled by a small number of platforms. A small-town activist could have the research capacity of a well-funded think tank with the help of a well-designed agent ecosystem.
Whether that vision of the future is supported by any significant political will is still up for debate. The past performance is not promising. Before attempting any meaningful response to social media, democratic governments watched the damage mount for about ten years, and those responses are still disputed and lacking. AI is moving more quickly than social media.
As this develops, it seems that the researchers closest to the issue are a little more concerned than their calm demeanor would indicate. The paper concludes with an emphasis on urgency rather than optimism, stating that unless democratic institutions are strengthened and repaired first, even the best-designed AI agent ecosystem will fail.
In the discussion of AI, that is the aspect that frequently gets overlooked. Broken institutions cannot be saved by technology. Anything that already exists is amplified. What is currently in place merits careful consideration.
