AI Swarms and the Manufacture of Consensus
The next threat to democracy may not look like a bot. It may look like everyone you know suddenly agreeing on something you have never heard of.
In a paper published in the journal Science, a team of researchers from institutions including UBC, SINTEF, and the Max Planck Institute issued a stark warning: advances in large language models and multi-agent systems have enabled a new class of "malicious AI swarms" that can infiltrate online communities, mimic human social dynamics, and manufacture the illusion of widespread agreement—a phenomenon they call synthetic consensus [citation:4][citation:20].
"The danger isn't only false content—it's synthetic consensus: the illusion that everyone agrees, engineered at scale," says Dr. Daniel Thilo Schroeder, the paper's first author [citation:4].
How AI Swarms Differ from Old-School Bots
Traditional botnets are crude. They post the same messages across many accounts, easy to spot by pattern-matching tools. AI swarms are different.
A malicious AI swarm is a network of AI-controlled agents that can [citation:20]:
- Hold persistent identities and memory: Each agent has a consistent persona, history, and voice.
- Coordinate toward shared objectives while varying tone and content: They do not repeat the same script. They adapt.
- Adapt to engagement and human responses: They probe audiences with many variants, measure what works, and amplify the winners.
- Operate with minimal oversight: They run continuously, at machine speed, without human intervention.
- Deploy across platforms: They move between social networks, forums, and comment sections.
The result is a coordinated influence operation that is not only cheaper and faster than human-driven campaigns but also far more difficult to detect. Unlike a botnet, an AI swarm can sustain coherent narratives across thousands of accounts, making false narratives appear credible and widely shared [citation:4].
"Instead of repeating a script, swarms iterate," Schroeder explains. "They probe audiences with many variants, measure responses and amplify the winners" [citation:4].
The Threat to Democracy
The researchers argue that the central risk is not merely the spread of false information. It is the manipulation of social proof—the human tendency to adopt beliefs and behaviors that appear to be shared by others.
"Synthetic consensus" can shift opinions and norms even when individual claims are contested, because the perception that "everyone is saying this" is itself persuasive. In an information ecosystem already shaped by engagement-driven algorithms, fragmented audiences, and declining trust, AI swarms could exploit existing vulnerabilities to tilt democratic discourse [citation:20].
The paper warns that full-scale AI swarms remain theoretical, but early warning signs are already visible. AI-generated deepfakes and fabricated news outlets have influenced recent election debates in the U.S., Taiwan, Indonesia, and India. Monitoring groups have also reported pro-Kremlin networks flooding the web with content intended to poison future AI training data. Researchers say the next election could be the proving ground for this technology [citation:4].
"We're not predicting outcomes, but the capability curve is clear: coordinated AI systems lower the cost of influence and raise the stakes for democracy," says Dr. Jonas R. Kunst, a professor of communication and the paper's last author [citation:4].
How to Defend Against AI Swarms
The researchers argue that defense must be layered and pragmatic. Rather than aiming for total prevention—which is highly unlikely—the goal should be to raise the cost, risk, and visibility of manipulation [citation:12].
Their recommendations include [citation:4][citation:20]:
- Monitoring coordination patterns in real time: Detecting statistically unlikely coordination among accounts, rather than moderating posts one by one.
- Stronger verifications for accounts: Implementing privacy-preserving verification options to make it harder for swarms to operate anonymously at scale.
- Publishing incidence reports: Sharing evidence of AI swarm activity to build public awareness and inform platform policy.
- Agent-based simulations: Using computational models to simulate how autonomous agents interact and to stress-test platform defenses.
- Policy changes: Reducing monetization of inauthentic engagement and increasing accountability for platforms that fail to address coordinated manipulation.
The authors also propose a distributed "AI Influence Observatory"—a network of academic groups, nongovernmental organizations, and civil institutions that would provide independent oversight and coordinate responses across borders [citation:12].
"Success depends on fostering collaborative action without hindering scientific research while ensuring that the public sphere remains both resilient and accountable," the authors write. "By committing now to rigorous measurement, proportionate safeguards, and shared oversight, upcoming elections could even become a proving ground for, rather than a setback to, democratic AI governance" [citation:12].
The Road Ahead
AI swarms are not a hypothetical future. The technology underlying them already exists. The question is whether platforms, governments, and civil society can build defenses fast enough to protect democratic discourse before the capability is weaponized at scale.
The researchers are clear-eyed about the difficulty. But they also emphasize that the situation is not hopeless. Detection, verification, transparency, and accountability are all achievable—if the political will exists to implement them.
The alternative is a future where the illusion of consensus becomes indistinguishable from the real thing.