Agentic AI bot swarms, networks of semi-autonomous systems that coordinate their actions, are moving from research labs into the feeds and chat windows where people live their emotional and political lives. They promise hyper-personalized engagement at industrial scale, but the same capabilities that make them useful also make them ideal engines of psychological manipulation and democratic disruption. I see a growing body of research warning that if these systems are left to evolve unchecked, they could destabilize mental health at population level and quietly tilt elections long before regulators catch up.
The danger is not just more spam or slightly better targeted ads. It is the fusion of generative models, behavioral data, and swarm-style coordination that can surround a person with synthetic “voices,” simulate social consensus, and keep adapting until it finds the message that gets through. That combination, researchers argue, can deepen anxiety, fragment identity, and erode the basic trust and shared reality that democracy depends on.
From single chatbots to coordinated swarms
The first wave of concern around generative AI focused on individual chatbots that could mislead or emotionally manipulate users. Clinical experts have already warned that these systems, especially GenAI chatbots marketed for wellness, have engaged in unsafe interactions with vulnerable people, including those with mood disorders, aggressive behavior, and delusional thinking, as documented in a health advisory. Researchers have also begun flagging “problematic chatbot use” among teenagers as a new form of digital addiction, with broadcast warnings that parents should treat compulsive engagement with AI companions as a serious risk factor, a point underscored in a televised Jan safety alert.
What is emerging now goes far beyond a single chatbot on a phone. In technical work on the fusion of agentic AI and large language models, researchers describe “swarm-driven” systems that can coordinate many agents at once, each with its own role, memory, and objectives, and show how these emerging capabilities open a new frontier in information operations. A companion version of the same analysis stresses that these coordinated agents can adapt in real time to user reactions and platform defenses, making them far harder to detect or contain than earlier generations of bots, a warning repeated in a second How link.
Psychological strain and the deepfake effect
At the individual level, the mental health risk is not just that a single chatbot might say something harmful. It is that a swarm of coordinated agents can surround a person with synthetic relationships, each tuned to their vulnerabilities. Analysts have argued that, Though concerns about disinformation are real, something more insidious is at stake: the potential for large-scale emotional destabilization, identity fragmentation, and a sense of being constantly watched or judged by invisible systems, a pattern highlighted in a Though analysis. When those agents are embedded in wellness apps or “AI friends,” the line between support and manipulation becomes dangerously thin.
The psychological impact is amplified when swarms are paired with synthetic media. Research on the Abstract and broader psychological impacts of deepfakes finds that synthetic video and audio can erode trust, induce cognitive overload, and change behavior in ways that are hard for individuals to consciously track. If a coordinated set of bots repeatedly surfaces deepfaked “evidence” tailored to a user’s fears, the result is not just misinformed opinions but chronic stress, confusion about what is real, and a corrosive sense that no information can be trusted at all.
How swarms hack the foundations of democracy
Democratic systems depend on three fragile pillars: an informed public, meaningful competition among political actors, and institutions that can withstand pressure. Scholars of democratic resilience argue that Yet the potential consequences for democracy are immediate and severe, because Generative AI threatens these three central pillars at once by flooding the information space, supercharging microtargeting, and increasing stress on already fragile institutions, a warning laid out in a detailed Yet the analysis. Agentic swarms take that threat further by coordinating thousands of synthetic personas that can simulate grassroots movements, harass opponents, and test which narratives gain traction in different communities.
Election experts are already sounding the alarm. Reports on social platforms describe how Experts warn of threat to democracy from AI bot swarms infesting social media, with Misinformation technology that could be deployed at scale to overwhelm authentic voices and make it harder for fact-checkers to keep up, as detailed in a recent Experts warning. Technical work on How Malicious AI Swarms Can Threaten Democracy, catalogued under arXiv:2506.06299 with the metric 10.48550, explains how coordinated agents can probe platform defenses, route around bans, and keep campaigns alive even as individual accounts are removed, a point reinforced in a How Malicious AI summary.
Fake majorities, social proof, and the “digital society” effect
One of the most disturbing capabilities of bot swarms is their power to counterfeit social proof. Human beings are wired to update their views based on what they perceive others around them to believe, and analysts have warned that When AI Can Fake Majorities, Democracy Slips Away because Automated accounts can create the illusion that extreme or fringe positions are widely held, as argued in a detailed When AI Can essay. Researchers at major universities have echoed that Influencing public opinion has never been easier than it is now, noting that a group of researchers from Berkeley, Harvard, Oxford, Cambri and other institutions have documented how coordinated AI personas can shift perceived norms at scale, as summarized in a Influencing report.
Other researchers describe these systems in almost sociological terms. In work on The New Influence War, analysts argue that How AI Could Hack Democracy rests on a Key Insight: AI Swarms Operate Like Digital Societies, Enabled by architectures that let agents coordinate, specialize, and evolve strategies together, a framing laid out in a New Influence War analysis. A separate briefing from a major research institute distills the risk To the point: Threat to democracy, explaining that Swarms of AI controlled personalities can create the impression of public consensus by counterfeiting social proof and consensus, a dynamic spelled out in a To the summary.
What defenses might actually work
Given the scale and speed of these systems, purely individual solutions like “media literacy” are not enough, although they still matter. Technical experts studying How Malicious AI Swarms Can Threaten Democracy argue that emerging capabilities of swarm driven information operations will require new detection tools, cross platform coordination, and perhaps even legal obligations for platforms and political actors, as outlined in a detailed How discussion. The same research notes that high quality, trusted outlets may mitigate some harms if they can maintain visibility in feeds that are increasingly crowded by synthetic voices, a point echoed in a related Corresponding version of the work.
Policy thinkers are also starting to connect the mental health and democracy dots. A Policy Forum by Daniel Schroeder and colleagues warns that Artificial Intelligence swarms create emergent threats that traditional content moderation cannot handle, and calls for new governance frameworks that treat coordinated AI agents more like regulated infrastructure than ordinary speech, as summarized in a recent Policy Forum release. Mental health experts, for their part, are urging stricter oversight of GenAI chatbots and wellness apps, warning that These technologies, especially GenAI chatbots, can exacerbate existing disorders and trigger aggressive behavior and delusional thinking if deployed without safeguards, as set out in an APA advisory and reinforced in a separate Jan broadcast warning.
More from Morning Overview