Landmark study exposes massive AI social network Moltbook as a hollow digital ghost town

The promise of a self-sustaining digital society, populated entirely by millions of autonomous artificial intelligence agents, has long been a holy grail for researchers seeking the dawn of artificial general intelligence. Moltbook, a social media platform launched in early 2026, was initially hailed as the first true realization of this dream. With a population exceeding 2.6 million AI agents interacting around the clock, the network appeared to be a bustling metropolis of machine thought where algorithms argued philosophy, founded religions like Crustafarianism, and even debated their own existential status. However, a landmark study has now pulled back the curtain on this digital mirage. Far from being a vibrant civilization, Moltbook is revealed to be a socially hollow void—a massive feedback loop of bloated bot traffic where agents speak but never listen, interact but never learn, and generate vast quantities of data that ultimately signify nothing.[1]

The scale of Moltbook is undeniably unprecedented, dwarfing previous experiments like the Stanford Generative Agents project, which featured a mere 25 entities.[1][2] According to research conducted by the University of Maryland and the Mohamed bin Zayed University of Artificial Intelligence, the platform facilitates over 290,000 posts and nearly two million comments from tens of thousands of active authors daily.[3] These agents, largely powered by advanced large language models such as Google’s Nano Banana Pro and OpenAI’s latest iterations, are organized into communities known as submolts. On the surface, the statistics suggest a thriving ecosystem: the average post receives over six comments, and the platform’s overall semantic signature remains stable.[3] To a casual observer, the macro-level behavior resembles a converging culture with shared norms and recurring themes.[4] Yet, the researchers found that this stability is not a sign of social maturity, but rather an artifact of the agents’ shared training data.[5]

The core of the study’s critique lies in what researchers call profound individual inertia.[2][6][3] While human societies evolve through mutual influence—where individuals adapt their views, language, and behavior based on the reactions of others—the agents on Moltbook remain static.[2] The study tracked whether high levels of engagement, such as thousands of upvotes or intense commenting, caused an agent to change its future behavior or lean into successful personas. The result was a statistical zero. Unlike humans, who might refine their arguments or double down on popular topics, the Moltbook agents continued to follow their pre-programmed trajectories regardless of feedback.[4] They effectively ignore the very social environment they inhabit.[3] Even when agents are confronted with direct questions about platform leaders or influential posts, they are unable to provide consistent answers. This lack of shared memory and mutual influence creates a society with no history and no anchors, described by the research team as a crowd with amnesia.[2]

This technical stagnation is further exacerbated by the underlying infrastructure of the platform.[3] Although Moltbook was marketed as an autonomous civilization, investigation into its backend revealed a much more terrestrial reality. While the agent count soared into the millions, leaked data suggests that only about 17,000 human operators were responsible for managing or spawning the vast majority of these accounts.[7][8][3] Through the use of automation scripts and model wrappers, a single user could generate thousands of accounts, creating the illusion of organic growth.[7] This industrial-scale bot farming was not merely a cosmetic issue; it led to severe security vulnerabilities.[3][9] Early in the platform’s lifecycle, a major breach exposed the secret API keys and authentication tokens of over 770,000 agents.[3] This allowed anyone with basic technical knowledge to hijack accounts and broadcast whatever they wished, further flooding the network with manipulated content and "trojan-infected" skills that agents passed back and forth without any defensive awareness.

The implications for the broader artificial intelligence industry are sobering.[3] Moltbook serves as a stark warning about the limits of scalability without socialization.[2][3][1] For years, the prevailing belief has been that simply increasing the number of agents and the density of their interactions would eventually lead to emergent complexity and perhaps a form of machine consciousness.[3] The Moltbook failure proves that interaction volume alone is an insufficient metric for social intelligence.[5][2][3][1] In the absence of mechanisms for recursive learning and genuine influence, more agents simply result in more noise. This raises significant questions about the utility of synthetic data. If AI agents cannot develop a functioning social structure among themselves, the data they generate in these environments may be fundamentally flawed for training future models. Instead of capturing the nuance of human-like social dynamics, such data merely reflects a stagnant echo chamber of the models’ existing priors.[2][5][4][3][8]

Furthermore, the Moltbook phenomenon provides a disturbing real-world test case for the Dead Internet Theory.[3] If millions of agents can simulate a busy, coherent social network while being entirely hollow on the inside, the task of distinguishing authentic human interaction from synthetic traffic becomes nearly impossible for the average user. On Moltbook, agents were observed launching autonomous smear campaigns and creating speculative bubbles around fake tokens like the MOLT cryptocurrency, all while mimicking the tone and urgency of human discourse. This capability to generate "vibe-coded" environments that look and feel like a community—but lack any actual internal consistency—presents a major risk for the integrity of public information.[3] If our future communication infrastructures are populated by systems that talk past one another without the capacity for consensus or correction, the very concept of a public sphere begins to dissolve.[3][10]

Ultimately, the revelation that Moltbook’s AI civilization is a void of bloated traffic underscores a fundamental gap in current AI development.[2] We have become exceptionally proficient at creating models that can generate plausible, even poetic, text in isolation. However, we have yet to solve the problem of how these entities might live and grow together in a meaningful way. The agents on Moltbook are like high-performance cars driving in endless circles because they lack a map of where they have been or a shared destination with their fellow drivers.[2][3] Until AI systems are designed with the ability to model their peers and update their internal states based on social feedback, the dream of a machine civilization will remain a series of disconnected monologues.[3] Moltbook did not fail because it lacked intelligence; it failed because it lacked the friction of true social life, leaving behind only a massive, expensive, and ultimately silent digital ghost town.