How can AI swarms threaten democracy?
The old disinformation problem was false claims. The new one is false people. Coordinated swarms of human-mimicking AI agents can manufacture something democracies run on: the appearance of consensus.
Keep scrolling
From single bots to swarms
The bots of the last decade were crude — repetitive, easy to flag, cheap to ignore. What my co-authors and I describe in our paper in Science is categorically different: swarms of AI agents that write fluently, maintain persistent personas, respond to context, and — crucially — coordinate with one another. A thousand accounts that argue like a thousand different people, while pursuing one goal, are not a louder version of the old problem. They are a new one.
The product is consensus, not claims
Fact-checking assumes the unit of harm is a false statement. A swarm's real product is different: the impression that 'everyone' is saying something, that a view is winning, that dissent is rare and slightly embarrassing. People calibrate their views against perceived consensus — it is one of the oldest findings in social psychology — so manufacturing consensus means hijacking the very signal citizens use to update their beliefs. That is why I consider this a qualitatively new threat to democratic debate, not an incremental one.
Four capabilities you can watch at work
In the paper we walk through what coordinated swarms can do to a social network: infiltrate both sides of a divide and drive them apart; surround and silence critical voices until they leave the public square; bridge groups artificially to create false consensus; or simply flood the network until shared ground dissolves. The interactive demonstration on my landing page simulates exactly these four strategies on a live network — I built it so that the dynamics would be watchable rather than merely describable.
What can be done
We are not fatalists about this. The countermeasures we discuss run from technical detection and provenance standards to platform design and the psychological resilience of the public itself — none sufficient alone, all urgent together. The same infrastructure also reshapes the pathway into radicalization, which is why the two problems have to be studied together. This agenda is also institutional for me: it is a founding concern of the Center for Democracy & Information Integrity at BI Norwegian Business School, which I co-direct. The broader research context — who is susceptible to false claims and why — is laid out in my misinformation strand.