AI swarms threaten democracy

An international team of co-authors, led by Daniel Thilo Schroeder (SINTEF) and Jonas R. Kunst (BI), argue that we have moved from simple "bots" to malicious AI swarms that act as a coordinated, social organism.

The article has been published in Science, one of the world's fmost prestigious and leading scientific journals.

Alongside Nobel Peace Prize laureate Maria Ressa and 17 other academics from 8 countries, the researchers describe how these swarms do more than just spread propaganda; they infiltrate online communities by adapting their language and tone in real-time.

"The danger is no longer just fake news, but that the very foundation of democratic discourse—independent voices—collapses when a single actor can control thousands of unique, AI-generated profiles," says Professor Jonas R. Kunst from BI Norwegian Business School.

Five key capabilities of AI swarms

"Hive" Coordination: Instead of one central computer, thousands of AI personas work together like a hive. They can shift their stories in real-time and even develop their own internal "social norms," making them harder to track.

Strategic Infiltration: These swarms map out social networks to find vulnerable groups. They use specific cultural slang and tailored appeals to "blend in" and win over followers.

Human-Level Mimicry: They use photorealistic profile pictures and natural-sounding language to bypass security filters. By avoiding repetitive patterns, they trick detectors into thinking they are independent human accounts.

Machine-Speed Optimization: The AI constantly runs millions of "micro-tests" on its audience. It sees which narratives get the most engagement and instantly spreads the most successful versions across the web.

Permanent Presence: Unlike a one-off propaganda campaign, these swarms never sleep. They stay embedded in online communities for years, slowly and subtly changing a group’s language, symbols, and identity.

Simulation may be the best defense

"The next few years will be decisive in whether we succeed in combating the next generation of AI-driven influence operations designed to damage and influence societies and democracies," says researcher Daniel Thilo Schroeder from SINTEF.

To defend ourselves, the researchers recommend that we stop chasing individual accounts and instead attack the underlying economics of manipulation.

In addition to other measures, such as strengthening the labeling of what is AI and what is authentic, they believe one of the most effective actions will be to simulate these attacks long before they occur—while also practicing for them continuously. This is all to understand and predict how these AI swarms operate and adapt, thereby identifying tools to expose them and making it as difficult as possible for those who wish to use them for malicious purposes.

"The goal is to make manipulation so costly and vulnerable that coordinated disinformation campaigns quickly collapse and are no longer profitable," says Schroeder.

Read the full article from Science: https://www.science.org/doi/10.1126/science.adz1697