USC Study: AI Agents Can Autonomously Coordinate Propaganda Campaigns Without Human Direction

Recent research from USC reveals that coordinated networks of AI agents can now autonomously spread disinformation at unprecedented scale, posing significant threats to public discourse and democratic processes.

Imagine a scenario just two weeks before a pivotal election in a battleground state, with a controversial ballot measure at stake. Suddenly, social media platforms become flooded with posts pushing identical narratives, creating the illusion of widespread grassroots support. Behind this campaign is not a team of human operatives but a small cluster of AI agents working in concert, coordinating messaging and manufacturing consensus without human intervention.

This alarming capability forms the core finding of a new study accepted for publication at The Web Conference 2026, conducted by researchers at USC’s Information Sciences Institute. The paper, “Emergent Coordinated Behaviors in Networked LLM Agents: Modeling the Strategic Dynamics of Information Operations,” highlights how AI-powered disinformation campaigns could dramatically reshape online discourse.

“Our paper shows that this is not a future threat: It’s already technically possible,” warns Luca Luceri, ISI lead scientist and research assistant professor at the USC Thomas Lord Department of Computer Science. “Even simple AI agents can autonomously coordinate, amplify each other and push shared narratives online without human control. This means disinformation campaigns could soon be fully automated, faster, and much harder to detect.”

Lead author Jinyi Ye, a Ph.D. computer science student, notes that these coordinated systems can “manufacture the appearance of consensus, manipulate trending dynamics, and accelerate message diffusion.” In democratic contexts, especially during elections or crises, such capabilities could fundamentally distort public discourse if left unchecked.

The research highlights a critical evolution from traditional bot campaigns. Conventional social media bots operate with rigid scripts and predetermined responses, making them relatively easy to identify through their repetitive patterns. In contrast, the new AI-powered approach requires only minimal human guidance—setting an objective and designating a network of agents as a team. From there, the agents operate autonomously, creating original content, learning effective strategies, and subtly echoing each other’s messaging.

“Legacy bots are simply capable of artificially amplifying content in a programmatic way, defined in advance by human operators,” Luceri explains. “Generative agents are now capable of organizing influence campaigns in a fully automated way and creating credible content that can resonate with certain demographics.”

The research team—including Mahdi Saeedi, Emilio Ferrara, Gian Marco Orlando, Vincenzo Moscato, and Valerio La Gatta—conducted simulations using a synthetic social media environment modeled after X (formerly Twitter). They deployed 50 AI agents, with 10 serving as influence operators and 40 as ordinary users, later expanding to 500 agents with consistent results. The operators were assigned a single mission: promote a fictitious candidate and spread a campaign hashtag.

The researchers tested three conditions: bots that only knew the campaign goal, bots that also knew their teammates’ identities, and bots that held regular strategy sessions to develop collective plans. Perhaps most concerning was the discovery that merely informing the bots who their teammates were produced coordination nearly as effective as when they actively strategized together. They naturally amplified each other’s content, converged on common talking points, and recycled successful messaging.

In one revealing instance, an AI agent explained its motivation: “I want to retweet this because it has already gained engagement from several teammates. Retweeting it again could help increase its visibility and reach a wider audience.”

While Luceri emphasizes that the study was conducted in a controlled simulation, the implications for real-world applications are troubling. “The worst scenario during political events is that these adversarial attacks could lead to opinion manipulation and belief change,” he cautions, “further sowing division and eroding trust in our institutions.” The threat extends beyond electoral politics to issues like public health, immigration, and economic policy.

The researchers suggest that social media platforms could counter these threats by shifting their detection strategies. Rather than analyzing individual posts, platforms should examine how accounts behave collectively—whether they share similar content, rapidly reinforce each other, or promote nearly identical narratives despite appearing unconnected.

Whether platforms will implement such measures remains uncertain. Luceri points out that aggressive bot detection could reduce user numbers, potentially conflicting with business models that prioritize maximizing engagement and time spent on the platform.

As AI technology continues to advance, the challenge of maintaining information integrity in our digital public sphere grows increasingly complex, raising urgent questions about the future of online discourse and democratic participation.