AI Swarms May Infiltrate Social Media to Manipulate Public Opinion at Scale

The Emergence of AI Swarms on Social Media

A new frontier in information warfare is taking shape as researchers identify the potential for “AI swarms”—autonomous networks of agents designed to mimic human behavior online. According to a commentary published January 22 in the journal Science, these swarms are capable of infiltrating digital communities to spread false narratives, harass dissenters, and manipulate public opinion at scale.

Unlike traditional bots that perform repetitive, easily detected tasks, these next-generation agents leverage large language models (LLMs) to adapt their personas and maintain identity over time. We talk about it as a kind of organism that is self-sufficient, that can coordinate itself, can learn, can adapt over time and, by that, specialize in exploiting human vulnerabilities, said Jonas Kunst, a professor of communication at the BI Norwegian Business School in Norway.

The potential for harm is significant. Researchers noted that these swarms could emulate an angry mob to target an individual with dissenting views and drive them off the platform. Jonas Kunst, a professor of communication at the BI Norwegian Business School in Norway, warned that human conformity makes the public particularly susceptible to such manipulation, as individuals often default to following the perceived wisdom of the herd.

Anthropic and the Risks of Recursive Self-Improvement

While AI agents threaten to influence human behavior, developers are simultaneously accelerating the internal capabilities of these models. On June 4, Anthropic released a report detailing how its Claude model now writes more than 80% of the code merged into its own production codebase. The firm frames this as the beginning of recursive self-improvement, where an AI designs its own successors without meaningful human input.

This trajectory has sparked debate regarding the future of human control. Anthropic’s internal data suggests that as of Q2 2026, engineers are merging eight times as much code per day as they were in 2024. Furthermore, on complex coding tasks, Claude succeeded 76% of the time in May 2026—a rise of 50 percentage points in six months. However, the company acknowledged that its reporting is based on internal data that has not been independently audited.

Critics remain skeptical of the narrative surrounding these developments. Cognitive scientist Gary Marcus described the report as a bait and switch, arguing that the data reflects faster coding under human direction rather than true autonomous improvement. Mathematician Noah Giansiracusa of Bentley University echoed this sentiment, stating, I don’t think it’s a genuine call to slow down.

The Philosophical and Practical Battle Over Artificial Wisdom

As AI capabilities expand, the industry is increasingly focused on the concept of artificial wisdom (AW). Proponents argue that current generative AI models, which rely on pattern-matching across vast datasets, lack the sound decision-making and insight required for true wisdom. The debate centers on whether such a quality can be engineered or if it is inherently tied to human embodiment.

This philosophical tension carries significant market stakes. Achieving artificial wisdom could move AI systems toward artificial general intelligence (AGI), potentially transforming the competitive landscape. As these models become more sophisticated, the challenge of alignment—keeping a system’s behavior tied to human intent—becomes more difficult.

The risks associated with these advancements are not theoretical. The International AI Safety Report, published in January 2025, defines the loss of control as a scenario where AI systems operate without any clear path for human intervention. With experts like Geoffrey Hinton estimating the probability of AI-caused human extinction at 10% to 20%, the industry continues to balance the race for compute-heavy development against the warnings of researchers who fear that rare misalignment in today’s models could keep growing more frequent but less understood until we lose control of them.