How malicious AI swarms can threaten democracy

The paper, in plain language

The argument

Written with colleagues spanning computer science, network science, psychology and journalism, this paper asks what changes when disinformation stops being produced by people. Our answer: the unit of harm shifts. Individual false claims can be checked; a swarm of fluent, persistent, mutually coordinating AI personas instead fabricates the social signal that citizens use to calibrate their own views — the sense of what everyone else thinks.

What makes swarms different

Earlier bot networks were repetitive and easy to flag. The agents we describe hold consistent personas over time, adapt to context, and coordinate strategy across thousands of accounts. That combination lets them do things no troll farm could sustain: infiltrate both sides of a divide and pull them apart, encircle and exhaust critical voices, or flood a network until shared ground dissolves. The interactive demonstration on this site's landing page simulates these dynamics on a live network.

What follows

We argue that no single countermeasure suffices: detection tools, provenance standards, platform design and public resilience have to move together, and the research infrastructure to study these systems barely exists. Building it is part of the agenda of the Center for Democracy & Information Integrity at BI Norwegian Business School, which I co-direct.